Non-Financial Factors and Financial Returns: The Impact of Linking ESG Metrics to Executive Compensation on Corporate Financial Performance

Abstract

Although the practice of linking Environmental, Social, and Governance (ESG) metrics to executive compensation (ESG compensation) has become increasingly common worldwide, consistent evidence of its economic consequences for corporate value remains limited. Drawing on agency theory and a sustainable governance perspective, this study examines how responsibility-oriented incentive mechanisms translate into corporate financial performance. Using textual data from a large sample of Chinese listed companies and employing the BERT deep learning model for empirical analysis, the results show that ESG compensation significantly improves subsequent financial performance. Further analysis reveals that this effect is primarily driven by incentives related to the environmental and social dimensions of compensation structures. In addition, ESG compensation enhances firms’ ESG rating performance and reduces rating divergence, thereby lowering stakeholders’ transaction costs. The moderating analysis indicates that managerial ability and financial slack both strengthen the positive effect of ESG compensation on financial performance. Overall, this study uncovers the internal mechanism through which ESG compensation promotes corporate value creation and clarifies its practical implications for sustainable corporate governance.

1. Introduction

Executive compensation is a crucial component of corporate governance, primarily aimed at aligning the interests of shareholders and managers and motivating managers to maximize shareholder value [1]. Traditionally, executive pay has been linked to financial metrics such as revenue growth, stock performance, and shareholder returns [2,3,4]. However, as global attention to ESG issues intensifies, integrating ESG metrics into executive compensation structures has emerged as a growing trend [5,6], described by some scholars as one of the most significant transformations in the domain of executive pay over the past decade [7]. Considering that effective compensation structures provide shareholders and corporations the opportunity to focus managerial actions on wealth creation, the key elements constituting effective compensation may have evolved over time [8]. Therefore, examining the efficacy of executive compensation structures and their impact on corporate financial performance requires a consideration of the growing importance of ESG factors, specifically whether introducing ESG compensation truly enhances financial performance [9].

Existing studies have examined the financial value of ESG activities from various economic perspectives, providing a solid theoretical foundation for analyzing the relationship between ESG compensation and financial performance. From the conservative viewpoint of neoclassical economics and agency theory, shareholders are the primary beneficiaries and residual claimants of the firm. When decision-making incorporates broader stakeholder interests, investment decisions may become suboptimal and managerial rent-seeking may increase, ultimately harming shareholder value [10]. Other research contends that ESG activities do not necessarily create new performance outcomes but rather redistribute existing ones. In this view, the costs incurred for ESG activities represent a form of “rent redistribution” among stakeholders [11].

Institutional theory regards ESG activities as a “dangerous placebo,” suggesting that many practices serve symbolic legitimacy rather than substantive actions [12]. In contrast, stakeholder theory increasingly posits that traditional profit-oriented business models and ESG-oriented strategies are not mutually exclusive. Some studies show that ESG practices can mitigate risk and accumulate reputational capital, providing firms with an “insurance mechanism” that enhances resilience and generates potential financial benefits during crises [13,14]. Furthermore, scholars emphasize that ESG considerations have become important criteria for investors in assessing firms and investment opportunities. Investors who value ESG may adjust their utility functions, granting higher risk tolerance and more favorable capital allocation to firms that demonstrate strong ESG performance [15,16]. Overall, this stream of research underscores that corporate financial success depends not only on relationships with shareholders but also on interactions with a broader set of stakeholders, a view that has become the dominant framework for explaining the link between ESG practices and financial performance.

As corporate governance continues to evolve, executive compensation contracts are shifting from a shareholder-oriented to a stakeholder-oriented paradigm [6]. Existing research has primarily focused on the mediating role of ESG ratings in information disclosure, while paying limited attention to the strategic function of compensation contracts as a micro-level governance tool within the ESG framework [17]. Although studies in developed economies have provided preliminary evidence on the governance value of ESG compensation, these findings are largely derived from mature markets. Their applicability in emerging economies, where institutional logic and market constraints differ significantly, remains uncertain [6,18].

Moreover, empirical research on ESG compensation is constrained by data availability. Due to limited disclosure, it is difficult to quantify the specific weight of ESG performance metrics in executive pay structures. Studies in developed markets often rely on executive compensation databases and use 0–1 type variables to indicate whether ESG targets are included in pay arrangements [6,19,20], while specialized databases are largely absent in emerging economies. China provides a particularly distinctive context, where ESG compensation practices are to a considerable extent shaped by administrative forces rather than market self-regulation or decentralized regulation. Exploring the governance effectiveness of ESG compensation under different policy constraints and legitimacy pressures can therefore reveal its contextual boundaries in emerging economies and enrich the existing literature.

This study uses China as a research setting and applies natural language processing with the BERT model to address the data gap in ESG compensation and systematically examine its impact on corporate financial performance. This approach responds to ongoing debates about its economic consequences and offers empirical evidence and methodological implications for emerging markets. Drawing on agency theory and institutional theory, this paper argues that ESG compensation serves both incentive and constraint functions. On the one hand, it optimizes contract design to alleviate agency conflicts and promote long-term decision orientation. On the other hand, it functions as an institutionalized responsibility-oriented incentive mechanism that enhances corporate legitimacy and external trust.

Empirical results show that incorporating ESG metrics into executive compensation significantly improves subsequent financial performance, with the environmental and social dimensions driving most of the effect. These responsibility-oriented objectives not only strengthen legitimacy but also convey credible responsibility signals to capital markets. In contrast, the governance dimension exerts a more long-term influence through improved internal control, reduced agency costs, and enhanced organizational resilience. Further analyses reveal that managerial ability and financial slack significantly strengthen the positive effect of ESG compensation on financial performance. High-ability managers can better allocate resources and implement ESG strategies, while sufficient financial slack provides firms with the flexibility to sustain ESG initiatives. Overall, the findings support the logic of agency theory that effective contract design promotes value creation and affirm the institutional theory proposition that legitimacy mechanisms drive sustainable organizational development. ESG compensation represents not only an innovative responsibility-oriented incentive but also a vital link between internal governance efficiency and external legitimacy. This study uncovers the mechanism through which ESG compensation fosters financial value creation via responsibility-oriented incentives, extends the theoretical boundary of executive incentives and sustainable corporate governance, and provides new theoretical and empirical evidence on the economic consequences of non-financial goals.

The contributions of this study are threefold. (1) It broadens the perspective of executive compensation research. Prior studies have made substantial progress in examining how different components of compensation structures—such as salary, bonuses, time-based and performance-based incentives, and payment instruments like cash and equity—affect incentive effectiveness [2,3,4]. However, limited attention has been given to the potential financial implications of innovative compensation designs. By analyzing ESG compensation as a novel pay structure, this study differentiates actionable managerial implications and offers a new analytical perspective on the relationship between ESG practices and financial performance.

(2) It provides new insights into how ESG compensation influences financial performance. The results show that integrating ESG metrics into compensation structures is positively associated with future corporate financial performance, primarily driven by the environmental and social dimensions. Firms with greater financial flexibility achieve stronger financial outcomes after adopting ESG compensation, offering empirical support for the slack resource theory. Moreover, ESG compensation is associated with greater consistency in ESG ratings, which is meaningful in light of increasing criticism of symbolic legitimacy. This suggests that authentic ESG practices enhance visibility and credibility among ESG-oriented investors, thereby generating financial benefits.

(3) It enriches cross-cultural comparative analyses of ESG practices. Since cultural systems differ in how they perceive legitimacy derived from ESG activities, the financial incentives and opportunities obtained from such practices also vary. In coordinated market economies such as those in Europe, ESG principles are often embedded within broader institutional norms and regarded as mandatory social expectations, so the direct financial incentives from ESG activities are limited. In contrast, in emerging economies like China, ESG has not yet become a universal legal or social requirement but remains a socially constructed expectation. Under such a context, ESG issues are more likely to be treated as managerial decisions rather than institutional mandates, which highlights the necessity of studying them from a managerial perspective. Methodologically, this study also introduces deep learning techniques to overcome limitations in ESG compensation data collection, offering new research ideas and methodological references for future studies.

2. Theoretical Analysis and Research Hypotheses

At present, research on ESG compensation remains in its early stage, and existing theories have not provided a clear ex-ante expectation regarding its relationship with financial performance. Since ESG compensation integrates the effectiveness concerns of traditional compensation incentives with the stakeholder values emphasized by ESG initiatives, discussions about its rationality involve both internal agency problems and external legitimacy issues. Based on these two economic perspectives, this study analyzes the relationship between ESG compensation and corporate financial performance as follows.

2.1. ESG Compensation as an Incentive Mechanism for Mitigating Agency Problems

The standard economic theory of executive compensation originates from the principal–agent model [21]. Within this framework, firms seek to design effective compensation strategies that attract, retain, and motivate executives to align their behavior with shareholders’ interests, thereby improving financial performance [18]. This mechanism essentially aims to address agency problems. As noted by Core et al. (1999) [22], an effective pay structure should establish an optimal economic contract between shareholders and management under transaction cost constraints to maximize economic value.

This perspective provides a theoretical basis for incorporating ESG metrics into compensation structures, as it helps address emerging agency problems [6]. These problems include inconsistencies between ESG and financial goals, trade-offs between short-term and long-term objectives, and the balance between transparency and financial outcomes. Without appropriate governance tools, managerial and shareholder objectives are difficult to reconcile. Flammer et al. (2019) [21] argue that if non-financial performance metrics provide additional information beyond financial indicators and reflect managerial effort, incorporating such metrics into compensation contracts can enhance contractual effectiveness, mitigate agency conflicts, and improve financial outcomes [23]. In this sense, establishing ESG-oriented compensation mechanisms not only responds to stakeholder expectations but also helps address the complex agency problems that arise from these expectations, aligning managerial behavior with broader stakeholder interests [19]. Hill and Jones (1992) [24] further point out that firms’ external relationships involve both explicit and implicit contracts with monitoring and constraint functions. Hence, ESG compensation helps reduce self-interested decisions made by executives. A key function of ESG pay is to compensate potential income loss when managers pursue ESG or other non-financial objectives at the expense of short-term financial performance.

From a functional perspective, compensation structures may also enhance managerial capability by developing better skills, processes, and information systems and by fostering a forward-looking managerial style that strengthens organizational responsiveness to external changes and crises [23]. Studies have shown that managing ESG resources is a complex task, and high-ability managers are more capable of integrating ESG issues within existing frameworks, leading to favorable financial outcomes [25]. As modern corporate governance increasingly emphasizes the welfare of diverse stakeholders, governance principles must focus on how inclusive decision-making can advance overall welfare and indirectly enhance shareholder value. Forward-looking managers are better positioned to interpret and implement these guiding frameworks, balancing multiple demands and fostering long-term sustainability, which is reflected in the integration and execution of ESG policies. ESG compensation provides the necessary incentives and support for managers to cultivate the skills required to handle complex challenges. Therefore, establishing ESG compensation mechanisms can strengthen internal capabilities, improve resource allocation efficiency, and ultimately enhance financial performance.

2.2. ESG Compensation as an Institutional Mechanism for Enhancing Legitimacy and Stakeholder Relationships

Institutional theory offers an alternative analytical logic. Prior studies generally regard stakeholder relationships as an important mediating mechanism through which ESG practices affect financial performance [17]. Evidence shows that firms with stronger ESG performance tend to exhibit higher investment efficiency, greater innovation capability, stronger reputational and social capital, and more effective decision-making processes [26,27]. From the perspective of the firm as a contract, a dense stakeholder network can lead to resource inefficiency if not well-coordinated, while ESG practices help balance stakeholder relationships and reduce such inefficiencies, thereby improving financial performance.

Barnett (2007) [11] distinguishes between foundational ESG initiatives and applied ESG initiatives, suggesting that the former better establish stable stakeholder relationships. Foundational ESG initiatives build overall credibility and reflect a firm’s long-term responsibility and judgment, while applied initiatives often target specific groups and are short-term in nature, sometimes at the expense of other stakeholders. Moreover, as societal expectations continue to rise—a phenomenon described as the Red Queen effect—firms that stagnate or fall behind in responding to stakeholder demands may lose competitive advantage.

In comparison, ESG compensation, as a typical foundational ESG initiative, involves long-term and structural adjustments that institutionalize responsibility fulfillment mechanisms and demonstrate altruistic motivation, thereby strengthening corporate credibility and stakeholder influence [19]. When facing disputes or crises, this stakeholder influence can transform into a reputation shield that stabilizes corporate performance through sustained relational capital [28]. Additionally, ESG compensation may help firms gain market recognition, attract capital support, and recruit talented employees [29], which further enhance financial outcomes. Therefore, the establishment of ESG compensation not only aligns with the logic of incentive compatibility within the firm but also conforms to the institutional orientation of building favorable stakeholder relationships.

Taken together, ESG compensation embodies both economic incentives and institutional constraints. Internally, performance-oriented responsibility incentives mitigate agency conflicts and promote long-term decision orientation. Externally, institutionalized incentives strengthen legitimacy, accumulate trust capital, and create sustainable reputational advantages. In emerging economies where policy guidance and legitimacy pressure coexist, ESG compensation functions as both a contractual innovation and a signaling mechanism, helping firms balance economic performance with social expectations. Based on this reasoning, this study proposes the following research hypothesis:

H1: 

Establishing ESG compensation has a positive effect on a firm’s future financial performance.

3. Research Design

3.1. Sample

In 2012, the China Securities Regulatory Commission (CSRC) released the Guidelines on Corporate Social Responsibility for Listed Companies (Draft for Comments), marking the institutionalization of corporate social responsibility (CSR) and sustainability disclosure in China. Since then, ESG-related disclosure by listed companies has become increasingly standardized, and information availability has improved significantly. Against this institutional background, this study selects A-share listed companies in China from 2012 to 2022 as the research sample. To ensure data reliability, the following screening procedures were applied. (1) Firms in the financial industry were excluded due to the unique characteristics of their capital structure and regulatory environment. (2) Firms under special treatment (ST or *ST) were excluded to avoid interference from financial distress or regulatory irregularities. (3) Samples with missing key variables were excluded to maintain the completeness of model estimation. (4) All continuous variables were winsorized at the 1st and 99th percentiles to mitigate the influence of extreme values and improve robustness. After these procedures, a total of 32,118 firm-year observations were obtained.

3.2. Model

This paper constructs the following model:

$${CFP}_{i,t+1}=\alpha +{\beta }_1{ESG\_pay}_{i,t}+{\beta }_{2}C{ontrols}_{i,t}+{\tau }_t+{\phi }_j+{\theta }_k+{\epsilon }_{i,t}$$

(1)

In Model (1), the dependent variable CFP_i,t+1 represents the company’s financial performance in period t + 1, including Roa, Roe, and Roic. The independent variable ESG_pay_i,t is a binary variable, set to 1 if the company has implemented an ESG compensation structure in period t, otherwise it is 0. Controls_i,t represents a vector of control variables for firm i in year t. Additionally, the model controls for time fixed effects to absorb impacts of economic cycles and other macroeconomic factors on strategic choices; industry fixed effects to absorb potential impacts of different industry risk exposures, market competition, and regulatory environments (The industry classification follows the Guidelines on Industry Classification for Listed Companies (2012 Revision); and regional fixed effects to account for potential impacts of economic, cultural, and regulatory differences, represented, respectively, by τ_t, φ_j, and θ_k.

3.3. Variable

3.3.1. Financial Performance

The measurement of corporate financial performance generally follows two frameworks, namely market-based measures and accounting-based measures. The former, such as Tobin’s Q and cumulative abnormal returns (CAR), are commonly used to evaluate market performance. However, Grewatsch and Kleindienst (2017) [30] suggest that market-based measures should be applied with caution when examining the relationship between ESG practices and financial performance. These indicators are grounded in the efficient market hypothesis, which assumes that market prices fully reflect all available information. This assumption has been widely questioned in the ESG context due to information asymmetry and the bounded rationality of investors, making the applicability of such measures controversial. To avoid inconsistency between theoretical assumptions and variable measurement, this study adopts accounting-based measures. Following prior studies [13,26,31], three indicators are selected: return on assets (Roa), return on equity (Roe), and return on invested capital (Roic). These measures have been extensively validated in the literature and demonstrate strong applicability. The corresponding data are obtained from the China Economic and Financial Research (CSMAR) database.

3.3.2. ESG Compensation

Research on ESG compensation is generally constrained by data availability limitations. Due to limited disclosure, it is often difficult for researchers to quantify the specific weight of ESG performance metrics in executive compensation. The mainstream literature typically uses a binary variable to identify whether firms have incorporated ESG-related performance constraints into their compensation contracts [6,19,20]. Following this approach, this study employs a binary variable to indicate whether a firm has introduced ESG metrics into its executive compensation arrangements. In the Chinese context, although there is no dedicated ESG compensation database, the gradual improvement of information disclosure standards has provided a foundation for identifying relevant information. Since the implementation of the Guidelines on the Content and Format of Information Disclosure by Companies Offering Securities to the Public No. 2—Annual Report (Revised in 2003), the disclosure of executive compensation by listed firms has become increasingly institutionalized. The 2007 revision further required companies to disclose details regarding decision-making procedures, determination criteria, and payment information. Although a unified disclosure template for ESG compensation has not yet been established, relevant descriptions are often found in the sections on compensation decision-making and performance evaluation in annual reports, making text-based identification feasible.

From a technical perspective, this study applies a BERT deep semantic model to identify the textual features related to ESG compensation in annual reports. As a significant advancement in natural language processing (NLP) [32], BERT demonstrates strong capability in semantic understanding and contextual capture, which substantially improves the precision and consistency of text recognition. Unlike traditional lexical analysis methods such as TF-IDF or Word2Vec, BERT is based on a Transformer architecture and employs a bidirectional attention mechanism to dynamically model the contextual relationships of each word within a sentence. This allows words to generate distinct semantic vector representations under different contexts. The mechanism not only identifies explicit keywords but also interprets implicit logical relations and semantic constraints, enabling more accurate detection of textual features that describe the linkage between executive compensation and ESG metrics.

This study uses the Chinese BERT model developed by Harbin Institute of Technology and completes model training and prediction in the Python 3.1 environment. The specific steps are as follows. First, a compensation keyword dictionary is constructed based on policy documents and corporate annual report texts. After manually reviewing and extracting common expressions related to compensation, Word2Vec and FastText algorithms are used for synonym expansion to enhance the dictionary’s coverage and representativeness. Second, the selected text samples are manually annotated and divided into training, validation, and test sets in proportions of 80%, 10%, and 10%, respectively. The annotation criteria follow the Stakeholder Capitalism Metrics: Towards Consistent Reporting and Sustainable Value Creation issued by the World Economic Forum (WEF). The research team received standardized training and labeled a sentence as a positive sample only when it explicitly indicated a linkage between executive compensation and ESG metrics; sentences that mentioned ESG content or compensation separately without explicit connection were labeled as negative samples.

During model training, the backpropagation algorithm was used to optimize the loss function, and parameter fine-tuning combined with multiple validation rounds was conducted to enhance the model’s generalization ability and prediction accuracy. After ten rounds of training and validation, the model achieved scores above 0.99 on all four performance metrics—Accuracy, Precision, Recall, and F1 score—demonstrating high recognition capability (detailed results are available upon request). Finally, the trained model was applied to the corporate annual report corpus. The identified results were manually reviewed, and sentence-level outputs were aggregated to the firm-year level. If a firm’s annual report contained at least one sentence explicitly indicating that executive compensation was linked to ESG metrics, the ESG_pay value for that year was set to 1; otherwise, it was set to 0.

3.3.3. Control

Regarding the control variables, this study first controls for the total executive cash compensation (Salary) to capture the incentive effect arising from the overall compensation level within the compensation contract, thereby better distinguishing variations caused by differences in compensation structure [33]. Second, financial leverage (Lev) and growth ability (Sgth) are included to reflect the influence of a firm’s current financial condition and development capacity on its future financial performance [34]. In addition, as prior studies suggest that corporate governance may affect firm performance [35], several governance-related variables are incorporated into the model, including board size (Board), the proportion of independent directors (Bi), CEO duality (Dual), ownership balance (Balance), and the ownership of the largest shareholder (Largest). To further account for firm characteristics that may influence the adoption of ESG compensation policies [36], the model also controls for firm size (Size), firm age (Age), engagement of Big Four accounting firms (Big4), and ownership type (SOE). Large firms are generally more likely to have resource advantages, while the degree of audit standardization and firm age may affect governance quality. Moreover, state-owned enterprises and non-state-owned enterprises often differ in their governance structures and goal orientations. The detailed measurement of all variables is presented in

Table A1. Variable definitions and measurements.

Variable	Definition
Roa	Net profit divided by total assets at year-end.
Roe	Net profit divided by shareholders’ equity at year-end.
Roic	(Net profit + financial expenses) divided by (total assets at year-end − current liabilities + notes payable + short-term borrowings + current portion of long-term debt).
ESG_pay	Dummy variable, equal to 1 if executive compensation is linked to ESG indicators in year t, and 0 otherwise; identified using the BERT model.
Size	Total assets at the end of year t (measured in CNY yuan); natural logarithm is taken.
Salary	Total cash compensation of top executives in year t (measured in CNY yuan); natural logarithm is taken.
Board	Natural logarithm of the total number of board members.
Lev	Total liabilities divided by total assets.
Bi	Number of independent directors divided by the total number of board members.
Largest	Shareholding percentage of the largest shareholder.
Sgth	Growth rate of operating revenue, calculated as (revenue in year t − revenue in year t − 1) divided by revenue in year t − 1.
Balance	Sum of the shareholding percentages of the 2nd to 5th largest shareholders divided by that of the largest shareholder.
Age	Number of years since the firm’s establishment; natural logarithm is taken.
Dual	Dummy variable, equal to 1 if the chairman and the CEO are the same person, and 0 otherwise.
Big4	Dummy variable, equal to 1 if the firm is audited by one of the Big Four accounting firms, and 0 otherwise.
Soe	Dummy variable, equal to 1 if the firm is a state-owned enterprise, and 0 otherwise.

Notes: Italicized items represent variable symbols used in the regression models, distinguishing them from abbreviations or descriptive text.

.

4. Empirical Results and Analysis

4.1. Descriptive Statistics

Table 1. Descriptive Statistics.

Variable	Sample Size	Mean	Standard Deviation	Minimum	Median	Maximum
ESG_pay	32,118	0.1661	0.3722	0.0000	0.0000	1.0000
Roa	32,118	0.0325	0.0674	−0.3352	0.0344	0.2070
Roe	32,118	0.0435	0.1599	−1.0045	0.0635	0.3458
Roic	32,118	0.0466	0.0858	−0.4464	0.0509	0.2725
Size	32,118	22.1760	1.3023	18.9964	21.9865	26.2090
Salary	32,118	15.3707	0.7162	13.4786	15.3509	17.3508
Board	32,118	2.2313	0.1738	1.7918	2.3026	2.7726
Lev	32,118	0.4111	0.2054	0.0535	0.3997	0.9433
Bi	32,118	37.5982	5.2499	33.3300	36.3600	57.1400
Largest	32,118	34.2727	14.6304	8.2600	32.0600	74.8200
Sgth	32,118	0.3732	0.9897	−0.7531	0.1289	6.9891
Balance	32,118	0.7528	0.6010	0.0322	0.5981	2.8342
Age	32,118	2.9314	0.3202	1.9459	2.9957	3.5264
Dual	32,118	0.3067	0.4611	0.0000	0.0000	1.0000
Big4	32,118	0.0583	0.2343	0.0000	0.0000	1.0000
Soe	32,118	0.3262	0.4688	0.0000	0.0000	1.0000

Notes: Italicized items represent variable symbols used in the regression models, distinguishing them from abbreviations or descriptive text.

presents the descriptive statistics of the main variables. The average adoption rate of ESG_pay is 16.61%. At the beginning of the sample period, 7.95% of firms implemented ESG compensation, increasing to 21.78% by the end. This upward trend reflects the growing attention to ESG practices and is consistent with the data reported in the Blue Book on ESG of Listed Central State-owned Enterprises (2022). For the three financial performance indicators, the mean values of Roa, Roe, and Roic are 0.0325, 0.0435, and 0.0466, respectively, with corresponding standard deviations of 0.0674, 0.1599, and 0.0858. The relatively high standard deviations of Roa and Roe suggest a considerable degree of variation in firm profitability. This pattern is not uncommon in firm-level financial data, particularly in the Chinese A-share market, where the sample includes both loss-making and high-growth firms. Such variation may reflect differences in firms’ profitability, risk exposure, and governance quality across industries. Regarding the control variables, the mean proportion of Bi is 37.60%, consistent with the minimum regulatory requirement of the China Securities Regulatory Commission. Other control variables fall within ranges commonly reported in the literature and are therefore not discussed further.

4.2. Baseline Test

Table 2. ESG Compensation and Firm Financial Performance.

Variable	Roat+1		Roet+1		Roict+1
	(1)	(2)	(3)	(4)	(5)	(6)
ESG_payt	0.0034 ***	0.0034 ***	0.0115 ***	0.0098 ***	0.0051 ***	0.0043 ***
	(3.3034)	(3.4058)	(4.6246)	(4.0498)	(3.8806)	(3.3016)
Sizet		0.0003		0.0050 ***		0.0005
		(0.7098)		(4.7465)		(0.8113)
Salaryt		0.0178 ***		0.0372 ***		0.0225 ***
		(27.2613)		(23.4211)		(26.4102)
Boardt		−0.0010		−0.0001		−0.0010
		(−0.3616)		(−0.0150)		(−0.2961)
Levt		−0.0831 ***		−0.1497 ***		−0.0500 ***
		(−38.3993)		(−28.3300)		(−17.6863)
Bit		0.0001		0.0002		0.0001
		(0.7607)		(1.0523)		(0.7801)
Largestt		0.0009 ***		0.0018 ***		0.0011 ***
		(25.7001)		(20.7713)		(23.7904)
Sgtht		0.0008 **		0.0041 ***		0.0009 *
		(1.9707)		(4.3732)		(1.7296)
Balancet		0.0088 ***		0.0154 ***		0.0101 ***
		(10.4470)		(7.5245)		(9.2362)
Aget		0.0006		−0.0007		0.0007
		(0.4389)		(−0.2272)		(0.4372)
Dualt		0.0014 *		0.0051 **		0.0020 *
		(1.7197)		(2.5507)		(1.8623)
Big4t		0.0018		0.0066 *		0.0023
		(1.1106)		(1.6719)		(1.0983)
Soet		−0.0012		−0.0010		−0.0020
		(−1.2478)		(−0.4342)		(−1.6069)
Constant	0.0400 ***	−0.2404 ***	0.0693 ***	−0.6236 ***	0.0609 ***	−0.3129 ***
	(6.9813)	(−18.6695)	(5.0418)	(−19.8391)	(8.3182)	(−18.5995)
Year	Yes	Yes	Yes	Yes	Yes	Yes
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Region	Yes	Yes	Yes	Yes	Yes	Yes
Observations	32,118	32,118	32,118	32,118	32,118	32,118
R-squared	0.0583	0.1576	0.0388	0.1077	0.0522	0.1124
F	16.50	45.32	10.77	29.25	14.69	30.67

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

presents the baseline regression results, indicating a positive impact of ESG compensation on future financial performance. In models without control variables (columns 1, 3, and 5), the estimated coefficients of ESG_pay on Roa, Roe, and Roic are 0.0034, 0.0115, and 0.0051, respectively, all significant at the 1% statistical level. After introducing control variables (columns 2, 4, and 6), the direction and significance level of the core explanatory variable remain unchanged, with coefficients of 0.0034, 0.0098, and 0.0043, respectively. These coefficients imply that the implementation of ESG compensation can increase the next year’s Roa by 0.34 percentage points, Roe by 0.98 percentage points, and Roic by 0.43 percentage points. These findings potentially confirm the value effect of incorporating ESG metrics into executive compensation, validating Hypothesis 1 (H1).

4.3. Robustness Tests

4.3.1. Two-Stage Least Squares and Instrumental Variables Method

The study may still face the endogeneity problem, particularly concerning the bidirectional causality between ESG compensation and financial performance. According to the slack resource theory, firms with better financial performance tend to have more available resources, which may facilitate greater investment in ESG activities. This implies that ESG practices and financial performance may be mutually causal. If the explanatory variable is endogenous, ordinary least squares (OLS) estimation may fail to accurately capture the true effect. Although the baseline model in this study incorporates lagged variables to mitigate reverse causality, the instrumental variable method and the two-stage least squares (2SLS) approach are employed to further identify the independent effect.

Regarding the choice of instrumental variables, this study draws on Ellul et al. (2024) [37], who used regional unemployment rates as a key instrument. Their research proposed that as employee unemployment risk changes, the board of directors adjusts the convex payoffs in CEO compensation schemes. This adjustment aims to encourage executives to avoid decisions that could negatively affect their careers while also taking on strategic risks that promote the firm’s long-term value growth. In this way, the adjusted compensation structure aligns managerial risk-taking with the firm’s long-term interests and enhances overall strategic execution. Their results confirm that boards consider employee unemployment risk when designing managerial incentive mechanisms. They also note that longer employee retention periods provide the board with both the opportunity and motivation to modify compensation schemes by introducing convex payoffs, which help reduce managerial risk aversion and career concerns while encouraging proactive engagement in strategic risks associated with long-term value creation. Based on this logic, regional unemployment rates are likely correlated with ESG-related indicators in executive compensation, as such indicators are often tied to the firm’s long-term value. Through performance-based incentives, ESG metrics may help alleviate executives’ career concerns related to ESG-related risks. At the same time, no empirical evidence suggests that regional unemployment rates directly affect corporate financial performance, satisfying the exclusion restriction condition.

The second instrumental variable follows another recent study [38], which finds that regional living costs significantly explain variations in executive compensation across China. The study highlights the “erosion effect” of high living costs on the real purchasing power of nominal compensation in first-tier cities. Contrary to the common assumption that first-tier cities naturally offer superior pay, their findings show that, after adjusting for living costs, cash compensation no longer provides an advantage, which may influence executive mobility trends. The study further reveals that in regions with higher living costs, boards tend to design more diversified compensation structures to strengthen executive retention incentives. Given the limited capacity of base salary to offset regional cost-of-living differences, diversified compensation mechanisms may help prevent executive outflow from high-cost areas such as Beijing, Shanghai, and Guangzhou. Therefore, firms in regions with higher living costs are more likely to link compensation structures to various business objectives, compensating for the limitations of monetary pay. This suggests a potential correlation between regional living costs and ESG compensation, while regional living costs are unlikely to have a direct effect on financial performance.

Urban registered unemployment rates are used to measure regional unemployment rates (EUR), and data are sourced from the China Population and Employment Statistics Yearbook. Regional minimum wage standards are used to measure regional living costs (RLC), with data obtained from the Chinese Research Data Services Platform (CNRDS). Empirical results from the 2SLS model using Roa as the dependent variable are reported in

Table 3. 2SLS.

Variable	First Stage	Second Stage
	ESG_payt	Roat+1
	(1)	(2)
EURt	−0.0166 ***
	(−2.7024)
RLCt	0.0895 ***
	(3.8794)
ESG_payt		0.0993 **
		(2.2622)
Sizet	0.0232 ***	−0.0020 *
	(9.1212)	(−1.8125)
Salaryt	−0.0068 *	0.0189 ***
	(−1.7404)	(22.8333)
Boardt	0.0659 ***	−0.0069
	(4.1717)	(−1.6097)
Levt	−0.0598 ***	−0.0763 ***
	(−4.6472)	(−20.8419)
Bit	0.0020 ***	−0.0001
	(4.1996)	(−1.0592)
Largestt	−0.0001	0.0009 ***
	(−0.5869)	(21.5477)
Sgtht	−0.0011	0.0007
	(−0.4736)	(1.5032)
Balancet	−0.0073	0.0096 ***
	(−1.4587)	(9.0806)
Aget	−0.0108	−0.0003
	(−1.4036)	(−0.1709)
Dualt	−0.0163 ***	0.0034 ***
	(−3.3257)	(2.8114)
Big4t	−0.0562 ***	0.0066 **
	(−5.9248)	(2.1238)
Soet	0.0733 ***	−0.0079 **
	(12.9961)	(−2.3155)
Constant	−1.1247 ***	−0.1884 ***
	(−6.1210)	(−6.6501)
Year	Yes	Yes
Industry	Yes	Yes
Region	Yes	Yes
N	28,921	28,921
Anderson canon. corr. LM statistic	21.267 ***
Kleibergen-Paap rk LM statistic	26.503 ***
Kleibergen-Paap rk Wald F statistic	13.284 [11.59]
Hansen J statistic	1.723 (p-value = 0.1893)
Sargan statistic	1.722 (p-value = 0.1895)

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

, while results for Roe and Roic are presented in

Table A2. 2SLS (continued).

Variable	First Stage	Second Stage
	ESG_payt	Roet+1	Roict+1
	(1)	(3)	(4)
EURt	−0.0166 ***
	(−2.7024)
RLCt	0.0895 ***
	(3.8794)
ESG_payt		0.2060 **	0.1078 *
		(1.9896)	(1.9436)
Sizet	0.0232 ***	0.0002	−0.0020
	(9.1212)	(0.0845)	(−1.4156)
Salaryt	−0.0068 *	0.0397 ***	0.0240 ***
	(−1.7404)	(20.3273)	(22.9123)
Boardt	0.0659 ***	−0.0119	−0.0075
	(4.1717)	(−1.1797)	(−1.3888)
Levt	−0.0598 ***	−0.1361 ***	−0.0433 ***
	(−4.6472)	(−15.7809)	(−9.3750)
Bit	0.0020 ***	−0.0002	−0.0001
	(4.1996)	(−0.7340)	(−0.8323)
Largestt	−0.0001	0.0018 ***	0.0011 ***
	(−0.5869)	(18.0339)	(20.6517)
Sgtht	−0.0011	0.0039 ***	0.0007
	(−0.4736)	(3.6572)	(1.2263)
Balancet	−0.0073	0.0165 ***	0.0110 ***
	(−1.4587)	(6.6320)	(8.2767)
Aget	−0.0108	−0.0014	−0.0004
	(−1.4036)	(−0.3635)	(−0.1815)
Dualt	−0.0163 ***	0.0093 ***	0.0042 ***
	(−3.3257)	(3.2849)	(2.7797)
Big4t	−0.0562 ***	0.0161 **	0.0074 *
	(−5.9248)	(2.2068)	(1.9049)
Soet	0.0733 ***	−0.0148 *	−0.0092 **
	(12.9961)	(−1.8299)	(−2.1315)
Constant	−1.1247 ***	−0.5171 ***	−0.2603 ***
	(−6.1210)	(−7.7415)	(−7.2755)
Year	Yes	Yes	Yes
Industry	Yes	Yes	Yes
Region	Yes	Yes	Yes
N	28,921	28,921	28,921
Anderson canon. corr. LM statistic	21.267 ***
Kleibergen-Paap rk LM statistic	26.503 ***
Kleibergen-Paap rk Wald F statistic	13.284 [11.59]
Hansen J statistic		0.532 (p-value = 0.4656)	2.112 (p-value = 0.1461)
Sargan statistic		0.536 (p-value = 0.4641)	2.131 (p-value = 0.1443)

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

. The results in the first column show that the coefficients for RLC and EUR are statistically significant and consistent with theoretical expectations. The Kleibergen-Paap rk Wald F-statistic exceeds the 15% Stock-Yogo critical value, indicating that the instruments are sufficiently strong. The Anderson canonical correlation LM statistic and the Kleibergen-Paap rk LM statistic have p-values below 0.1, indicating that the model passes the identification test. The Hansen J statistic and Sargan statistic have p-values greater than 0.1, indicating that the model passes the exogeneity test. In the second-stage regression, the coefficient for ESG_pay remains significant, supporting the foundational hypothesis.

4.3.2. Entropy Balancing Robustness Test

Despite incorporating numerous control variables, along with fixed effects for time, industry, and region to minimize biases and absorb the impact of both macro and micro factors, there remains a concern about unquantifiable omitted variables. To address this issue further, this study employs entropy balancing as a robustness test. Specifically, the study first estimates weightings by using all available control variables and then applies these weights to the regression model to control for potential biases. The regression results, presented in

Table 4. Entropy balancing test.

Variable	Roat+1	Roet+1	Roict+1
	(1)	(2)	(3)
ESG_payt	0.0017 *	0.0053 **	0.0022 **
	(1.9194)	(2.3967)	(1.9678)
Sizet	0.0017 ***	0.0078 ***	0.0021 ***
	(3.0359)	(5.1997)	(2.7716)
Salaryt	0.0164 ***	0.0343 ***	0.0202 ***
	(17.3103)	(15.1087)	(16.9943)
Boardt	0.0054	0.0164 *	0.0077 *
	(1.5433)	(1.8837)	(1.7319)
Levt	−0.0914 ***	−0.1610 ***	−0.0592 ***
	(−26.8960)	(−15.1245)	(−13.0492)
Bit	0.0001	0.0005 **	0.0001
	(0.9449)	(2.1969)	(1.2047)
Largestt	0.0007 ***	0.0014 ***	0.0008 ***
	(15.7280)	(12.6643)	(14.7234)
Sgtht	0.0010 *	0.0054 ***	0.0016 **
	(1.9137)	(4.0898)	(2.2172)
Balancet	0.0047 ***	0.0086 ***	0.0055 ***
	(4.0199)	(2.8322)	(3.5751)
Aget	−0.0011	−0.0031	−0.0018
	(−0.6715)	(−0.7811)	(−0.8601)
Dualt	0.0011	0.0033	0.0018
	(1.0354)	(1.2529)	(1.2555)
Big4t	−0.0031	−0.0054	−0.0042 *
	(−1.5894)	(−1.1376)	(−1.7475)
Soet	−0.0023 *	−0.0037	−0.0037 **
	(−1.8714)	(−1.1633)	(−2.3515)
Constant	−0.2444 ***	−0.6664 ***	−0.3128 ***
	(−15.7526)	(−15.7866)	(−15.4916)
Year	Yes	Yes	Yes
Industry	Yes	Yes	Yes
Region	Yes	Yes	Yes
Observations	32,118	32,118	32,118
R-squared	0.1867	0.1205	0.1288
F	28.89	50.19	23.78

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

, show a decrease in the significance levels of the key variable ESG_pay compared to the baseline tests, but the sign of the coefficients remains consistent and is significant at least at the 10% level.

4.3.3. Propensity Score Matching Analysis

Given that the implementation of ESG compensation in firms may systematically be influenced by observable characteristics such as company size and governance level, ordinary least squares (OLS) may conflate the effects of policy with inherent corporate traits. Employing Propensity Score Matching (PSM) facilitates the construction of a counterfactual control group, aiding in the isolation of corporate characteristic differences from the estimation results. Therefore, this study further conducts robustness checks using PSM. Based on all observable characteristics (control variables), control groups are constructed through year-by-year 1:1 nearest neighbor matching without replacement. The balance test results are displayed in

Table A3. Balance test.

Variable	Before Propensity Score Matching
	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022
Age	0.0304	0.0074	0.0001	−0.0157	−0.0282 *	−0.0395 ***	−0.0335 ***	−0.0284 **	−0.0396 ***	−0.0432 ***	−0.0617 ***
Balance	0.0715	0.0554	0.1068 ***	0.0952 ***	0.0593 *	0.1079 ***	0.1496 ***	0.1341 ***	0.1218 ***	0.1024 ***	0.1216 ***
Bi	0.6693 *	−0.2068	0.2924	0.5800 *	0.3221	0.1394	0.1217	−0.0476	0.3204	0.0678	0.1258
Big4	0.0163	−0.0291	−0.0184	−0.0175	−0.0199	−0.0338 **	−0.0218 *	−0.0172	0.0013	0.0048	−0.0087
Board	−0.0479 ***	−0.0360 ***	−0.0513 ***	−0.0536 ***	−0.0428 ***	−0.0458 ***	−0.0391 ***	−0.0377 ***	−0.0471 ***	−0.0411 ***	−0.0416 ***
Dual	0.0461	0.0327	0.0714 ***	0.1072 ***	0.0793 ***	0.1061 ***	0.1246 ***	0.1042 ***	0.1093 ***	0.1119 ***	0.1226 ***
Largest	−3.2045 **	−4.2998 ***	−2.9619 ***	−3.4456 ***	−1.7188 *	−2.6785 ***	−3.6536 ***	−3.1709 ***	−3.0586 ***	−2.2193 ***	−1.9763 ***
Lev	−0.0187	−0.0227	−0.0408 ***	−0.0365 ***	−0.0490 ***	−0.0480 ***	−0.0350 ***	−0.0358 ***	−0.0353 ***	−0.0280 ***	−0.0425 ***
Salary	−0.0223	−0.0777 *	−0.0848 *	−0.0670	−0.0161	−0.0132	−0.0111	−0.0075	0.0249	−0.0076	−0.0506 **
Sgth	0.1063	0.0773	0.1347 **	0.1669 ***	0.0786	0.0918 *	0.0723 *	0.1195 ***	0.0920 ***	0.0726 **	0.0755 ***
Variable	After propensity score matching
	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022
Age	−0.0170	0.0317	0.0275	0.0039	0.0102	0.0079	−0.0030	0.0078	0.0066	−0.0064	−0.0108
Balance	−0.0577	0.0221	0.0310	0.0103	0.0147	0.0064	−0.0041	0.0179	0.0107	0.0034	0.0082
Bi	0.3238	0.1777	−0.1532	−0.1060	−0.3459	0.3323	−0.0895	−0.0920	0.0263	−0.2469	0.0235
Big4	0.0000	−0.0087	−0.0077	−0.0033	0.0141	−0.0069	0.0000	−0.0139	−0.0122	0.0000	−0.0104
Board	−0.0129	−0.0037	−0.0010	−0.0075	0.0090	−0.0120	−0.0067	−0.0008	−0.0067	0.0026	−0.0022
Dual	−0.0122	−0.0043	0.0077	0.0230	0.0056	−0.0253	−0.0019	0.0052	0.0031	−0.0129	0.0000
Largest	1.4250	−1.5488	−2.3790 *	−0.7118	−0.7067	−1.0087	−1.2618	−0.1987	−0.9341	−0.2629	−0.0018
Lev	−0.0101	−0.0054	−0.0043	−0.0102	0.0048	0.0018	0.0031	−0.0034	−0.0004	0.0000	−0.0098
Salary	−0.0362	−0.0158	−0.0168	−0.0095	0.0300	0.0007	0.0272	−0.0383	0.0049	−0.0013	0.0021
Sgth	0.0469	0.0329	0.0683	0.0814	0.0108	0.0023	0.0413	0.0359	0.0120	−0.0165	−0.0095

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

, indicating that potential biases in all control variables have been addressed. The regression results, as shown in

Table 5. PSM sample robustness test.

Variable	Roat+1		Roet+1		Roict+1
	(1)	(2)	(3)	(4)	(5)	(6)
ESG_payt	0.002 **	0.002 **	0.007 ***	0.007 ***	0.003 **	0.003 **
	(2.005)	(2.059)	(2.652)	(2.720)	(2.172)	(2.165)
Sizet		0.001 *		0.007 ***		0.002 *
		(1.956)		(4.131)		(1.740)
Salaryt		0.017 ***		0.036 ***		0.022 ***
		(17.008)		(14.287)		(16.551)
Boardt		0.005		0.017		0.008
		(1.249)		(1.593)		(1.426)
Levt		−0.092 ***		−0.165 ***		−0.060 ***
		(−26.103)		(−18.824)		(−13.270)
Bit		0.000 **		0.001 **		0.000 **
		(2.069)		(2.512)		(2.103)
Largestt		0.001 ***		0.001 ***		0.001 ***
		(11.795)		(9.765)		(10.871)
Sgtht		0.001		0.003 **		0.001
		(1.513)		(2.026)		(1.325)
Balancet		0.004 ***		0.008 **		0.004 **
		(2.913)		(2.229)		(2.381)
Aget		0.001		−0.002		0.000
		(0.308)		(−0.451)		(0.120)
Dualt		0.001		0.002		0.001
		(0.749)		(0.707)		(0.740)
Big4t		−0.002		−0.005		−0.003
		(−0.687)		(−0.779)		(−0.799)
Soet		−0.002 *		−0.002		−0.004 *
		(−1.649)		(−0.521)		(−1.911)
Constant	0.035 ***	−0.263 ***	0.057 ***	−0.682 ***	0.056 ***	−0.336 ***
	(4.180)	(−13.355)	(2.782)	(−13.905)	(5.254)	(−13.217)
Year	Yes	Yes	Yes	Yes	Yes	Yes
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Region	Yes	Yes	Yes	Yes	Yes	Yes
Observations	10,508	10,508	10,508	10,508	10,508	10,508
R-squared	0.082	0.193	0.052	0.127	0.071	0.137
F	8.183	19.71	5.015	11.99	7.015	13.09

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

, reveal no substantial changes in the coefficients of the core variables.

4.3.4. Addressing Misleading Disclosures

The accuracy of core indicators may be compromised by firms’ selective disclosure practices. To mitigate this issue, this study employs data samples following the enactment of two key regulatory documents, re-performing regression analyses to diminish the potential impact of misleading disclosures. Firstly, Guideline No. 2 (Revised in 2017) builds upon its 2007 revision, which required listed companies to disclose the decision-making processes, bases for determination, and actual payments of senior management compensation. The 2017 revision further mandated that disclosures should be straightforward, truthful, and sincere, ensuring that information conveyed to investors is accurate and encompasses details about the decision processes, bases for determination, and actual payments concerning senior management compensation. Secondly, the 2018 revision of the “Corporate Governance Code” by the CSRC explicitly requires listed companies to lawfully disclose detailed ESG-related information, establishing a clear framework for ESG disclosure. This revision implements a “disclose or explain” policy, enhancing specific requirements for ESG information disclosure. The issuance of these documents likely promoted the standardization and authenticity of disclosures related to ESG compensation. Particularly, the “Corporate Governance Code (Revised in 2018)” not only established a clear framework for ESG disclosure for the first time but also specified disclosure requirements, fundamentally improving the standardization of ESG information among listed companies and limiting the scope for misleading disclosures regarding ESG compensation. Based on this, the study re-analyzes samples from 2018 onward to reduce the effects of potential misleading disclosures. The results, as shown in

Table 6. Addressing misleading disclosures.

Variable	Roat+1		Roet+1		Roict+1
	(1)	(2)	(3)	(4)	(5)	(6)
ESG_payt	0.003 **	0.003 **	0.012 ***	0.010 ***	0.005 ***	0.004 **
	(2.469)	(2.439)	(3.558)	(3.019)	(3.044)	(2.497)
Sizet		0.001 **		0.008 ***		0.002 **
		(2.017)		(5.494)		(2.034)
Salaryt		0.019 ***		0.037 ***		0.023 ***
		(19.846)		(16.451)		(19.110)
Boardt		0.003		0.008		0.005
		(0.711)		(0.908)		(0.966)
Levt		−0.087 ***		−0.183 ***		−0.060 ***
		(−28.534)		(−24.761)		(−14.992)
Bit		0.000 *		0.001 **		0.000 *
		(1.855)		(2.183)		(1.959)
Largestt		0.001 ***		0.002 ***		0.001 ***
		(21.220)		(17.326)		(19.797)
Sgtht		0.000		0.004 ***		0.001
		(0.720)		(3.031)		(0.846)
Balancet		0.011 ***		0.020 ***		0.013 ***
		(9.308)		(7.051)		(8.689)
Aget		−0.001		−0.004		−0.001
		(−0.412)		(−0.896)		(−0.613)
Dualt		0.001		0.006 **		0.002
		(1.117)		(2.104)		(1.373)
Big4t		0.000		0.002		−0.000
		(0.171)		(0.391)		(−0.031)
Soet		0.000		0.004		−0.000
		(0.350)		(1.236)		(−0.011)
Constant	0.032 ***	−0.300 ***	0.057 ***	−0.742 ***	0.052 ***	−0.386 ***
	(3.843)	(−16.116)	(2.804)	(−16.431)	(4.823)	(−15.911)
Year	Yes	Yes	Yes	Yes	Yes	Yes
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Region	Yes	Yes	Yes	Yes	Yes	Yes
Observations	18,385	18,385	18,385	18,385	18,385	18,385
R-squared	0.073	0.173	0.054	0.132	0.067	0.133
F	12.68	30.31	9.119	22.08	11.49	22.16

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

, indicate that the coefficients of the core variables remain consistent in terms of sign and significance level.

4.3.5. Inferring Causal Relationships Through Within-Group Variability

The baseline model follows the control methodologies used in prior research [6], incorporating fixed effects for time, industry, and region. While this approach is well-suited for scenarios with minimal variability in independent variables, such as changes in compensation policies, it may still allow for some irrelevant variability between groups. To ensure that the variations in the dependent variable genuinely stem from the influences of the independent variables, this study introduces a fixed effects model at the individual level. This model strictly infers causality using changes within individuals over time, thereby rigorously controlling for irrelevant intergroup variability. The regression results, as displayed in

Table 7. Inferring causal relationships using within-group variability.

Variable	Roat+1	Roet+1	Roict+1
	(1)	(2)	(3)
ESG_payt	0.0044 ***	0.0103 ***	0.0050 **
	(3.0110)	(2.6703)	(2.5643)
Sizet	−0.0183 ***	−0.0373 ***	−0.0220 ***
	(−19.8513)	(−15.4757)	(−18.0864)
Salaryt	0.0097 ***	0.0237 ***	0.0120 ***
	(9.4738)	(8.7941)	(8.8449)
Boardt	−0.0029	−0.0142	−0.0027
	(−0.6419)	(−1.2159)	(−0.4559)
Levt	−0.0087 **	−0.0530 ***	0.0334 ***
	(−2.5365)	(−5.9234)	(7.4039)
Bit	0.0000	0.0000	−0.0000
	(0.0654)	(0.0960)	(−0.0914)
Largestt	0.0009 ***	0.0020 ***	0.0010 ***
	(11.9372)	(10.4268)	(10.6399)
Sgtht	0.0027 ***	0.0081 ***	0.0034 ***
	(6.6798)	(7.6064)	(6.3505)
Balancet	0.0097 ***	0.0182 ***	0.0108 ***
	(6.5489)	(4.6780)	(5.4783)
Aget	−0.0089	0.0254	−0.0037
	(−1.2486)	(1.3658)	(−0.3929)
Dualt	0.0020 *	0.0046	0.0029 *
	(1.7676)	(1.5139)	(1.9101)
Big4t	0.0105 ***	0.0250 ***	0.0120 ***
	(3.1968)	(2.8917)	(2.7536)
Soet	−0.0099 ***	−0.0387 ***	−0.0152 ***
	(−4.1698)	(−6.1890)	(−4.8254)
Constant	0.2866 ***	0.4246 ***	0.3175 ***
	(9.8196)	(5.5489)	(8.2156)
Year	Yes	Yes	Yes
Firm	Yes	Yes	Yes
Observations	32,118	32,118	32,118
R-squared	0.0559	0.0431	0.0517
F	70.19	53.39	64.68

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

, show that the outcomes for all core variables do not exhibit substantial changes.

4.4. Moderating Effect Analysis

4.4.1. Managerial Ability

This study further identifies the moderating factors affecting the efficacy of ESG compensation. Existing research often focuses on the moderating mechanisms of manager characteristics on ESG practices, with a comprehensive review by Aguinis & Glavas (2012) [39] confirming the significant moderating effect of managerial characteristics on the relationship between ESG and financial performance. Other literature suggests that under the framework of fiduciary duties, executives face multiple constraints in allocating ESG resources, whereas those with exceptional managerial ability can effectively integrate ESG goals with corporate strategy, enhancing financial value [40]. Representing a comprehensive capacity that encompasses resource allocation, strategic decision-making, and organizational mobilization, managerial ability is not only linked to controlling agency costs but is also a core driver of value creation. From an agency theory perspective, managers with high ability have dual motives to promote ESG compensation: one is to mitigate the myopic incentives in traditional compensation structures; the other is to reduce the personal utility losses incurred from pursuing non-financial goals.

In conclusion, it can be inferred that managerial ability might play a significant moderating role in the positive relationship between ESG compensation and financial performance. In terms of variable construction, traditionally, scholars rely on decision theory and the resource-based view, using proxies such as manager reputation, compensation, and the company’s abnormal earnings to indirectly assess managerial capability, which may be affected by individual characteristics and market environmental changes. In contrast, methods based on production efficiency theory provide a more scientific approach to assessing managerial ability. This study adopts the measurement framework proposed by Demerjian et al. (2012) [41], using the Data Envelopment Analysis (DEA) combined with the Tobit regression model to calculate managerial ability. Specifically, the DEA method is first used to calculate the industry’s optimal efficiency score for sample companies, Maxθ:

$${Max\theta }_{j,t}={\displaystyle \frac{{TS}_{j,i,t}}{w_1{NFA}_{j,i,t}+w_2{IA}_{j,i,t}+w_3{GW}_{j,i,t}+w_4{RDE}_{j,i,t}+w_5{CGS}_{j,i,t}+w_6{SAE}_{j,i,t}+w_7{NOLE}_{j,i,t}}}$$

(2)

where the output indicator uses the company’s total sales (TS), and the input indicators include net fixed assets (NFA), intangible assets (IA), goodwill (GW), R&D expenditure (RDE), cost of goods sold (CGS), selling and administrative expenses (SAE), and net operating lease expenses (NOLE). The weights w_k (k = 1, 2, …, 7) for each indicator are automatically determined through the DEA model’s linear programming. Then, using the industry’s optimal score as a benchmark, the company’s score θ is calculated by comparing its relative distance from the industry’s best performance, Maxθ, and is standardized, with scores ranging from 0 to 1. A score of 1 indicates full efficiency, with lower scores showing a greater gap between the company’s management efficiency and the industry frontier. Subsequently, a Tobit model is established:

$${\theta }_{i,t}={\beta }_0+{\beta }_1{Size}_{i,t}+{\beta }_2{Market}_{i, t}+{\beta }_3{FCF}_{i,t}+{\beta }_4{Age}_{i,t}+{\beta }_5{HHI}_{i,t}+{\beta }_6{FC}_{i,t}+{\tau }_t+{\phi }_j+{\epsilon }_{i,t}$$

(3)

Control variables include company size (Size), market share (Market), free cash flow (FCF), years listed (Age), degree of diversification (HHI), and the presence of overseas subsidiaries (FC), thereby eliminating factors that might affect the company’s production efficiency at the corporate level. The model also controls for industry and year fixed effects, thus excluding the impact of industry-specific factors on efficiency. The calculated residuals represent managerial ability (Ability), with values increasing from weakest to strongest.

4.4.2. Financial Slack

The level of financial slack may also influence the quality and sustainability of corporate ESG practices. Based on the slack resources theory, organizational redundant resources provide a strategic buffer for ESG practices: firms with excess cash flow or collateralizable assets can overcome the constraints of short-term survival pressures to implement ESG initiatives with long-term value. Research by Duque-Grisales & Aguilera-Caracuel (2021) [42] suggests that firms under financial constraints are more likely to adopt symbolic strategies, while those with idle resources can drive substantive change. This may stem from three mechanisms of financial slack: (1) reducing the crowding-out effect of debt contracts on free cash flow, (2) building risk tolerance to support long-cycle projects, and (3) mitigating the tendency toward investment myopia in agency conflicts. Consequently, this paper proposes that the effects of ESG compensation may be more pronounced in firms with high levels of financial slack. In terms of variable construction, financial slack is defined as Slack = (Cash and cash equivalents/Total assets) + (Net value of collateralizable assets/Total assets).

Regression results as shown in

Table 8. Moderation effects analysis.

Variable	Roat+1	Roet+1	Roict+1	Roat+1	Roet+1	Roict+1
	(1)	(2)	(3)	(4)	(5)	(6)
ESG_payt	0.0033 ***	0.0099 ***	0.0042 ***	0.0036 ***	0.0091 ***	0.0043 ***
	(3.2983)	(4.0886)	(3.2167)	(3.6897)	(3.9437)	(3.4127)
Abilityt	0.0770 ***	0.1425 ***	0.0952 ***
	(31.1898)	(23.8239)	(29.5967)
ESG_payt × Abilityt	0.0099 *	0.0257 **	0.0117 *
	(1.8937)	(2.0217)	(1.7117)
Slackt				0.0667 ***	0.1477 ***	0.0796 ***
				(22.2241)	(20.5421)	(20.3853)
ESG_payt × Slackt				0.0086 **	0.0157 *	0.0112 **
				(2.2977)	(1.7481)	(2.2921)
Constant	−0.2807 ***	−0.6982 ***	−0.3616 ***	−0.2960 ***	−0.7269 ***	−0.3797 ***
	(−21.3907)	(−21.9478)	(−21.1562)	(−23.0690)	(−23.6561)	(−22.7691)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes	Yes	Yes
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Region	Yes	Yes	Yes	Yes	Yes	Yes
Observations	29,521	29,521	29,521	31,526	31,526	31,526
R-squared	0.1926	0.1293	0.1452	0.1781	0.1232	0.1309
F	53.10	33.07	37.83	51.14	33.17	35.55

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

, columns 1–3 report findings with managerial ability (Ability) as a moderating variable, and columns 4–6 with financial slack (Slack) as a moderating variable. Results indicate that both Ability and Slack play a significant positive moderating role between ESG compensation and financial performance.

5. Additional Analysis

5.1. Heterogeneity Across the Three Pillars: A Subdivision Perspective

Considering previous research suggesting that different ESG dimensions may have varying levels of significance for financial performance [31], this study explores how subdividing ESG compensation into specific dimensions may differentially impact financial outcomes. To categorize statements related to ESG compensation identified through manual review, each statement that involves a specific dimension is marked accordingly: E_pay, S_pay, or G_pay is set to 1 if related, otherwise 0.

The regression results presented in

Table 9. Analysis of ESG Compensation by dimension.

Variable	Roat+1	Roet+1	Roict+1
	(1)	(2)	(3)
E_payt	0.0027 **	0.0055 *	0.0029 *
	(2.2386)	(1.8708)	(1.8662)
S_payt	0.0042 **	0.0081 *	0.0051 **
	(2.3593)	(1.8562)	(2.2122)
G_payt	−0.0004	0.0042	−0.0002
	(−0.2277)	(1.0188)	(−0.0868)
Constant	−0.2403 ***	−0.6241 ***	−0.3130 ***
	(−18.6571)	(−19.8503)	(−18.5997)
Controls	Yes	Yes	Yes
Year	Yes	Yes	Yes
Industry	Yes	Yes	Yes
Region	Yes	Yes	Yes
Observations	32,118	32,118	32,118
R-squared	0.1576	0.1076	0.1124
F	44.67	28.78	30.22

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

show that compensation variables related to E_pay and S_pay aspects are significantly positive; however, G_pay related compensation variable, while positive, does not reach statistical significance. Results for Roe and Roic as dependent variables (shown in columns 4 and 6, respectively) did not fundamentally change. These outcomes suggest that the motivational effects of ESG compensation currently stem mainly from environmental and social factors, with governance-related compensation having a relatively smaller impact on financial performance. Additionally, the coefficient for the S_pay is higher than that for E_pay. This aligns with findings by Sandberg et al. (2023) [43], who noted that among the three ESG components, social elements are the most crucial for enhancing corporate efficiency and financial performance. The observed differences might also be explained by the transmission mechanisms of institutional pressures, as environmental and social indicators typically respond directly to the explicit demands of external stakeholders, generate immediate financial returns by enhancing market legitimacy, and consumers show significantly higher sensitivity to commitments to environmental protection and social responsibility than to governance structures. These make E and S indicators more likely to translate into competitive advantages. The role of governance indicators, however, often reflects in long-term operational efficiency improvements, with value realization requiring the restructuring of organizational norms, hence their lack of significant correlation with financial performance.

5.2. Compensation Structure and the “Fog” of ESG Ratings

As ESG factors become a key criterion for evaluating corporate operations and investment opportunities, many companies are striving to improve their ESG ratings. However, the breadth and complexity of ESG, involving frequently overlapping concepts and often imprecisely measurable indicators, constrain the accurate assessment of corporate efforts in ESG. Furthermore, variations in ESG rating standards and preferences among different rating agencies add to this complexity, limiting precise evaluations of a company’s overall ESG capabilities [44]. These factors often lead to contradictory views in studies using ESG ratings as research variables. The potential inconsistencies in ESG ratings provided by different agencies make it difficult for companies to determine which ESG actions will meet market expectations or gain recognition. Flammer et al. (2019) [21] suggest that incorporating ESG metrics into executive compensation serves as a signal of high seller quality and non-opportunistic behavior. Thus, ESG compensation might be seen by stakeholders as a substantive measure that not only responds to external expectations but more importantly, establishes an internal driving mechanism for fulfilling ESG responsibilities, thereby appearing more sincere and potentially endorsing the accuracy of ESG ratings and bridging rating discrepancies. Based on this analysis, this paper further explores the impact of ESG compensation on rating disagreements.

This study utilizes ESG rating data provided by six agencies: Bloomberg, FTSE, Wind, Sino, SynTao, and Susall. To ensure comparability between data, the ratings were first standardized, then the degree of disagreement among different rating agencies’ ESG scores was measured by calculating the standard deviation (Diff1) and the range (Diff2). The regression results are shown in

Table 10. ESG Compensation, ESG Ratings, and Discrepancies.

Variable	Diff1t	Diff2t	Bloombergt	FTSEt	Windt	Sinot	SynTaot	Susallt
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
ESG_payt	−0.0239 **	−0.0381 **	0.0536 ***	0.1069 **	0.0778 ***	0.0529 ***	0.0237	−0.0586
	(−2.2261)	(−2.0348)	(2.6787)	(2.2948)	(5.1262)	(3.0718)	(0.9939)	(−1.5757)
Constant	−2.1160 ***	−8.9625 ***	−3.5209 ***	−9.1378 ***	0.4141 *	−4.1793 ***	−4.1279 ***	−12.8743 ***
	(−14.3340)	(−34.8403)	(−13.0260)	(−11.6317)	(1.8950)	(−17.6676)	(−11.3036)	(−21.3596)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Industry	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Region	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	25,009	25,009	7626	2509	17,239	25,009	8068	3562
R-squared	0.2775	0.3975	0.4442	0.4055	0.2117	0.1914	0.2521	0.5090
F	74.65	128.2	49.99	14.73	36.78	46.01	21.07	30.51

Notes: t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. p-values are based on two-tailed tests. Standard errors are robust and clustered at the firm level. Italicized terms denote variable symbols.

, with columns 3 to 8 displaying the regression outcomes with each ESG rating as the dependent variable. In columns 3–7, ESG_pay is positively correlated with ESG ratings, with results in columns 3–6 significant at the 1% level. Columns 1–2 report results with Diff1 and Diff2 as dependent variables, consistently showing a significant negative correlation, suggesting that implementing ESG compensation helps unify discrepancies in ESG ratings. This finding indicates that linking ESG metrics with executive compensation may help distinguish the firm from those that may only comply superficially or “symbolically legitimate,” thereby providing investors and other stakeholders with a more consistent and clear signal of ESG practices.

6. Discussion

Existing literature has not definitively established the relationship between ESG compensation and corporate financial performance, with uncertainties primarily arising from two aspects: the internal decision-making context regarding such compensation structures remains unclear, and the broader market and regulatory framework’s acceptance and evaluation of this pay model are still in question. Addressing this issue, this paper analyzes the potential impacts from both the internal decision-making and external institutional environments using a large-sample analysis constructed with BERT deep learning and NLP. This study aims to clarify the feasibility of ESG compensation in the increasingly ESG-focused corporate world.

Our findings suggest a positive correlation between the establishment of ESG compensation and future corporate financial performance, primarily driven by environmental (E) and social (S) components. These dimensions often align with broad social expectations and ethical standards, thereby enhancing a company’s perceived legitimacy. For example, implementing environmental initiatives or engaging in social causes can quickly improve market recognition and legitimacy. In contrast, the effects of governance (G) appear more indirect. Rather than shaping external perceptions, governance contributes to long-term stability by improving internal structures and operational processes. Theoretically, sound governance should enhance financial performance by reducing agency costs and strengthening internal control. Yet in practice, such benefits may be limited when senior managers have vested interests in maintaining existing arrangements, which constrains their incentives to pursue substantial structural reforms. Beyond these internal and structural mechanisms, external market reactions also reinforce the value of ESG compensation. The market responds positively, as reflected in higher institutional ratings and more consistent evaluations. These improvements help distinguish firms that genuinely commit to substantive ESG initiatives from those engaging only symbolically, thus providing credible signals to investors and other stakeholders.

The analysis of moderating effects reveals that managerial ability and financial slack play crucial roles in successfully translating ESG compensation into significant financial outcomes. Specifically, managerial ability is essential for effectively mobilizing and allocating organizational resources and seizing opportunities related to ESG. These capabilities not only accelerate the implementation of ESG measures but also ensure that these actions are reflected in the company’s financial performance. Additionally, financial slack provides the necessary resource buffer for executing ESG strategies, allowing companies to sustain ESG efforts under lower financial risk.

Despite its contributions, this study has limitations and areas for improvement. A key issue in compensation research is whether the established structures help companies achieve returns that exceed marginal costs. Due to data constraints, our analysis was based on dummy variables, thus limiting our ability to quantify specific economic effects. Future research could delve deeper into quantifying the specific benefits of ESG compensation, thereby providing more definitive evidence for its economic rationality and addressing concerns about “rent redistribution” more clearly.