Promoting or Hindering: The Impact of ESG Rating Differences on Energy Enterprises’ Green Transformation—A Causal Test from Double Machine-Learning Algorithms

Abstract

There is a lack of comprehensive evaluation on the impact of ESG rating differences on the green transformation of energy enterprises in the transition era. This study leverages data from companies listed on the Shanghai Stock Exchange in China, applying double machine-learning algorithms to precisely estimate the causal relationship between variations in ESG ratings and the green transition efficiency of energy companies. The research shows that the difference in ESG ratings of third-party rating agencies significantly promotes the efficiency of green transformation of energy enterprises. This paper also studies the influencing factors of this effect: First, ESG rating differences significantly promote the improvement of green transition efficiency of energy enterprises; Second, the positive effect is more pronounced in energy companies with more balanced board structures. Finally, energy companies with high capital market attention can also contribute to this positive impact. Through the mechanism test, this paper finds that enterprise green innovation is an important mechanism for ESG rating divergence to positively promote the efficiency of energy enterprises’ green transformation. Furthermore, this paper analyzes the impact of ESG rating on enterprises from the perspective of market cognition and short-term behavior, which provides a new perspective for analyzing the practice of enterprises pursuing long-term transformation. The study also calls for a more sober reflection on the trend toward ESG in society.

1. Introduction

Against the backdrop of global environmental deterioration, increasingly severe environmental issues have drawn widespread attention from various sectors around the world [1]. Currently, the pursuit of green and sustainable development has become a core issue of our time [2]. In China, all sectors of society have deeply realized the necessity of abandoning the previous development model characterized by “high energy consumption and high emissions” [3], shifting instead towards promoting a green transformation of the economy to achieve higher-quality growth [4]. In September 2020, China set clear goals for reaching carbon peaking by 2030 and achieving carbon neutrality by 2060 [5]. Under the new normal, green innovation transformation plays a more significant role than comprehensive energy innovation in driving economic growth and reducing energy consumption [4]. As the basic unit of economic development, the green transformation of enterprises is particularly critical, especially for energy companies [6]. Researchers have noted in their scientific reflections that energy companies hold a strategic and undeniable role in the operation and further development of the modern economy [7]. Therefore, the ability of energy companies to adopt green development practices and successfully undergo transformation is crucial for reshaping economic growth patterns, achieving the “dual carbon” targets, and fostering high-quality economic development [8].

In the current environment, environmental, social, and governance (ESG) issues have become a focal point in recent years [9]. The term ESG represents an emerging investment and business concept that refers to the incorporation of environmental protection, social responsibility, and corporate governance into investment decisions and business practices [10]. It has become a key indicator for measuring a company’s sustainability performance [11]. With the increasing demand for ESG-related information from stakeholders, rating agencies have become essential players in this domain [12]. However, accurately assessing corporate ESG performance presents considerable challenges [13]. In practice, discrepancies in ESG ratings often arise due to conflicts of interest or differences in evaluation methodologies among rating agencies [14]. These rating discrepancies leave stakeholders confused when making decisions based on ESG scores and have attracted widespread attention from the academic community [15]. It is noteworthy that discrepancies in ESG ratings are a factor that stakeholders must consider carefully when making decisions based on ESG scores [16]. Particularly in corporate decision-making processes, ESG ratings directly impact stock prices, earnings, and market performance [17,18]. Therefore, whether companies will adopt strategic behaviors to reduce ESG rating discrepancies to cater to rating agencies’ preferences and the potential consequences of these actions are crucial topics for further research in both academia and business practice.

In this study, we analyze whether discrepancies in ESG ratings affect the efficiency of green transformation in energy companies and explore the specific mechanisms of this impact. The starting point of the research is the insufficient attention paid to the ESG ratings of energy companies in the existing literature, limiting the possibility of an in-depth understanding of the ESG performance of energy companies. Most existing studies focus on the association between ESG ratings and corporate financial performance and their importance in measuring a company’s ability to achieve sustainable development [19,20,21,22,23,24], but they neglect the ratings themselves and their potential to influence carbon reduction efforts in energy companies. Some scholars have begun to focus on the inherent issues of ESG ratings, expressing concerns about their lack of transparency [25], unclear standards [26], lack of consistency [14], and scale bias [27], pointing out that these issues may lead to stakeholder misunderstandings of companies’ carbon reduction commitments [28]. The scale bias of ESG ratings marginalizes small and medium-sized energy companies in their carbon reduction efforts [29], despite these companies playing an equally indispensable role in the overall energy transition [30,31,32]. Moreover, in the limited studies on ESG rating discrepancies, scholars primarily focus on exploring their causes, such as differences in the selection of evaluation indicators [33], measurement methods [34], weight allocation [35], and subjective judgments [36].

While existing research provides valuable insights into the causes of ESG rating discrepancies [36,37], the effects of these discrepancies on corporate value remain relatively underexplored. Only a few studies have examined the impact of ESG rating discrepancies from the perspectives of stock market reactions [38], financial constraints [36,39], and greenwashing [40]. Regarding ESG discrepancies and energy company green transformation efficiency, existing research indicates that ESG investments can effectively secure the future economic safety of energy companies in the context of green transformation [6]. Ref. [41] pointed out that the growing interest of investors in ESG encourages companies to seek more sustainable development paths to meet ESG standards and investor demands [41]. Ref. [36] also noted that significant differences among ESG rating providers indicate that investors and stakeholders must exercise caution when relying on these ratings for decision-making, particularly in energy efficiency projects that require consistent evaluation [36]. Ref. [42] mentioned that ESG ratings may not significantly improve companies’ sustainable development behavior and could mislead stakeholders [42]. Clément et al. (2022) [25] further argued that ESG scores lack design features to measure sustainability concepts such as temporality, impact, resource management, and interconnectedness, failing to adequately capture sustainability [25].

In conclusion, the literature lacks sufficient evidence on the consequences of ESG rating discrepancies, particularly regarding their impact on the sustainability of energy companies. However, more scholars believe that ESG has a positive role in the green transformation of energy companies. Therefore, linking ESG performance and its discrepancies with the green transformation efficiency of energy companies provides a new perspective for analyzing the role of ESG performance in the sustainable development of energy companies. Ref. [43] found that differences in ESG practices across industries highlight the need for tailored approaches to corporate green transformation [43]. Ref. [44] also argued that investors’ growing focus on ESG factors has led to increased scrutiny of ESG differences between companies, prompting them to strengthen their sustainability disclosures and accelerate green transformation initiatives to meet stakeholder expectations [44]. Energy companies play a crucial role in society’s overall carbon reduction process, and ESG ratings are a market-based supervisory measure for the sustainable transformation of companies in the energy sector. However, there is still a gap in research on how ESG ratings affect the green transformation of companies in the energy industry, and no consensus has been reached on the relationship between the two. Whether energy companies can address ESG rating discrepancies by improving green transformation efficiency is still an unresolved question in academia. Therefore, under the guidance of the concept of sustainable development, our research aims to fill this gap by examining the role of ESG rating discrepancies in the sustainable development of energy companies from the perspective of green transformation efficiency, providing new insights into the economic consequences of ESG rating discrepancies.

To this end, we empirically examine whether and how ESG rating discrepancies affect the green transformation efficiency of energy companies. Using a sample of listed companies in China, we find strong evidence that ESG rating discrepancies significantly promote the improvement of energy companies’ green transformation efficiency. From the internal and external perspectives of corporate governance, we further find that the positive impact of ESG rating discrepancies on energy companies’ green transformation efficiency is less pronounced in companies with more balanced board structures and higher capital market attention. In addition, by exploring the role of green innovation in this supervisory process, this paper also provides an in-depth analysis of how ESG rating discrepancies influence energy companies’ green transformation efficiency [45]. Finally, by examining the impact of ESG rating discrepancies on the uncertainty and financial performance of energy companies, we reveal their overall impact on the performance of energy companies. Overall, our research reveals the significant impact of ESG ratings on the sustainability outcomes of energy companies.

2. Theoretical Analysis and Hypothesis Formulation

To achieve sustainable development, energy companies have come under widespread ESG evaluations [46]. Whether ESG discrepancies affect the efficiency of energy companies’ green transitions and contribute to their high-quality development is a question worth analyzing. Stakeholder theory suggests that companies should not only meet shareholder interests but also consider the needs of a broad range of stakeholders [47]. ESG ratings, as a key indicator of corporate sustainability performance, reflect how well companies perform in environmental protection, social responsibility, and good governance [48]. However, the efficiency of green transitions in energy companies, which is a more implicit assessment metric, is often not fully incorporated into ESG rating systems. Improving green transition efficiency means energy companies can use energy more effectively in their production processes, reducing waste and carbon emissions, thus reflecting their capacity to respond to climate change and dual carbon policies. Nevertheless, this efficiency improvement is essential for energy companies to progress efficiently towards a carbon-neutral future. For most energy companies, zero carbon emissions in production are impossible [49]. Only by improving green transition efficiency can companies advance their carbon neutrality process.

In the context of pursuing sustainable development, ESG ratings, as a vital measure of corporate sustainability performance, have been widely applied [50]. ESG ratings cover environmental, social, and governance dimensions and aim to comprehensively reflect corporate sustainability [51]. Prior research has demonstrated a significant positive correlation between favorable ESG ratings and stock price premiums [52]. This indicates that managers are strongly motivated to improve their companies’ ESG ratings to gain positive recognition from the capital market and attract investors [53]. However, discrepancies in ESG ratings can lead to mistrust and skepticism from the capital market, prompting managers to take steps to reduce rating discrepancies to gain broader investor recognition [38]. Therefore, energy companies may aim to improve their sustainability performance by narrowing ESG rating discrepancies, with green transitions being a key pathway to achieving corporate sustainability. Through more efficient green transitions, companies can more effectively enhance resource utilization efficiency, reduce energy consumption, and promote sustainable development.

In summary, driven by ESG rating discrepancies, energy companies will improve their green transition efficiency to use energy more effectively, thereby pursuing enhanced sustainability performance. Therefore, we propose Hypothesis 1:

H1. 

ESG rating discrepancies significantly promote the improvement of green transition efficiency in energy companies.

Based on the above analysis, we believe that the extent of this positive effect varies across energy companies with different board structures, thus affecting the improvement of green transition efficiency. We will discuss this from three aspects.

First, we consider the impact of the proportion of independent directors on decision-making in energy companies. Independent directors, as a key component of corporate governance, bring more objective perspectives and higher professional expertise. They can objectively assess strategic decisions, such as formulating and implementing plans to improve green transition efficiency, thus improving the supervision and decision-making quality of the board [54]. In companies with a higher proportion of independent directors, ESG factors tend to carry more weight [55], as independent directors are more inclined to consider long-term interests, corporate image, and social responsibility [56], which leads to greater attention to ESG issues. Consequently, they more actively supervise environmental policies, social responsibilities, and governance practices. Moreover, independent directors often demand that energy companies take action to improve green transition efficiency to meet social and environmental expectations [57]. Therefore, in cases of significant ESG rating discrepancies, independent boards push companies to focus more on substantial sustainability metrics such as green transition efficiency [58], directing resources toward these areas rather than merely striving for short-term improvements in ESG scores. As a result, the positive impact of ESG discrepancies on improving green transition efficiency should be more evident in energy companies with a higher proportion of independent directors.

Second, gender diversity on boards is another important factor. Research shows that increasing the proportion of women on boards often leads to more cautious and forward-thinking decision-making, with greater attention to long-term sustainability rather than short-term gains [59]. Therefore, gender-diverse boards tend to be more innovative and inclusive, capable of considering a broader range of stakeholder needs and expectations and bringing different perspectives and approaches to decision-making [60]. This diversity helps avoid groupthink and improve decision quality [61]. In this context, boards with a higher gender ratio are more inclined to take actions that align with ESG standards rather than merely pursuing short-term profit maximization. They focus more on substantive sustainability metrics like green transition efficiency, as this aligns with environmental and social responsibilities and long-term economic interests [62]. However, in companies with a lower proportion of female directors, the lack of diversity may lead to boards being overly focused on short-term performance evaluations, with less emphasis on long-term projects. Consequently, resources may be allocated to issues yielding immediate returns, reducing investment in green transition efficiency. Therefore, the positive impact of ESG discrepancies on improving green transition efficiency should be more pronounced in energy companies with higher gender diversity on their boards.

Lastly, the influence of the largest director on corporate strategy and operations is crucial. The largest director typically refers to the board member with the most voting power or influence, holding dominance in decision-making. If the largest director’s influence is too great, it can lead to a dictatorship-like governance structure where decisions are overly concentrated, lacking diversity and balance. In this case, companies may not fully weigh various interests or engage in collective discussions. Individual directors, being irrational, tend to pursue immediate returns and short-term interests, often prioritizing their own interests over those of other stakeholders [63]. In these companies, when faced with ESG rating discrepancies, managers may allocate resources to actions that reduce discrepancies to gain market recognition, boost stock prices, and enhance personal interests. As a result, green transition efficiency, which requires long-term investment, might be overlooked. According to this analysis, for energy companies with a lower proportion of shares held by the largest director, the positive impact of ESG discrepancies on improving green transition efficiency should be more evident.

Thus, by examining the proportion of independent directors, gender diversity, and the shareholding ratio of the largest director in governance structures, we propose the second hypothesis:

H2. 

In energy companies with more balanced governance structures (higher proportions of independent and female directors and lower largest director shareholding), the positive impact of ESG rating discrepancies on green transition efficiency is more pronounced.

The capital market’s attention refers to the level of interest and responsiveness that the market shows toward a particular company or industry. For the capital market, companies with higher visibility tend to attract more investor attention, thereby securing more funding and lower financing costs [64]. Therefore, the extent of this positive effect varies across energy companies depending on their visibility in the capital market.

When capital market attention is high, companies’ actions will come under broader public and investor scrutiny, and investors generally have higher expectations for companies’ environmental actions and climate change responses [65]. This creates pressure for companies to not only focus on their ESG performance but also take substantive emission reduction measures. If a company’s actions do not match its ESG performance claims, the company may face public skepticism and lose investor trust, risking stock price declines and divestment [66]. In this situation, high market attention acts as a soft supervisory mechanism [67], making companies more cautious about balancing ESG rating discrepancies and green transition efficiency improvements. Aware of the high level of market attention, companies understand that besides improving ESG performance, substantive emission reductions are equally important. Thus, energy companies will focus more on improving green transition efficiency to reduce carbon emissions and minimize negative environmental impacts, thereby avoiding public and investor scrutiny.

In summary, we propose the third hypothesis:

H3. 

The positive impact of ESG rating discrepancies is more significant in energy companies with higher capital market attention.

3. Research Design

3.1. Sample Selection and Data Sources

The sample in this paper consists of A-share companies listed on the Shanghai and Shenzhen stock exchanges from 2007 to 2022. The ESG data of the companies comes from the ratings of Wind, China Securities Index (CSI), FTSE Russell, SynTao Green Finance, Minglang, and Bloomberg. All other data are obtained from the CSMAR database. For the selection of energy sector companies, this paper follows the 2012 industry classification standards of the China Securities Regulatory Commission (CSRC), with sub-sector categories including coal mining and washing, oil and gas extraction, petroleum refining, coking and nuclear fuel processing, electricity and thermal power production and supply, and gas production and supply. Referring to existing literature, the following steps were taken to process the raw data: (1) Financial companies were excluded since they typically do not engage in industrial activities. (2) Companies labeled as ST or PT during the sample period were excluded for the corresponding year. (3) Due to the contagious nature of missing values in R Studio R 4.0.2, samples with missing data were removed.

3.2. Variable Quantification

3.2.1. ESG Rating Divergence

The explanatory variable ESG_dif represents ESG rating divergence. As ESG concepts have continued to develop and deepen, numerous ESG rating agencies have emerged, each with its unique calculation indicators and measurement methods. This paper selects six representative ESG rating agencies: CSI, FTSE Russell, Minglang, Bloomberg, SynTao Green Finance, and Wind, and calculates the standard deviation of their ratings to obtain ESG rating divergence data. CSI and Wind ratings are divided into 9 levels, and SynTao Green Finance ratings are divided into 10 levels, so these were assigned values ranging from 1 to 9 and 0 to 9, respectively. To make the scores of the other four agencies comparable, this paper processed them accordingly. Referring to Sustainable Investing with ESG Rating Uncertainty for data processing [38], Minglang ratings, for example, are divided into 19 enhanced levels from low to high, with the lowest level assigned a value of 0 points, the second-lowest level assigned 1 point, and so on, with the highest level assigned 18 points. The values were then multiplied by 9/18 to scale the scores down to 0 to 9. For Runling Global, which has 7 levels and lacks C and CC levels compared to CSI and Wind, the ratings would be 2 points lower. In other words, when a company is rated CCC, CSI and Wind would assign a value of 3. For FTSE Russell and Bloomberg, the specific ESG rating scores were processed by taking 10% of Bloomberg’s ESG score and 200% of FTSE Russell’s score as sample data. If only one ESG rating is available, the divergence value is 0. After processing, the ESG ratings of the seven agencies all range from 0 to 9, meeting the comparability requirements.

3.2.2. Corporate Green Transformation

This paper draws on the research method from How does digitalization affect the green transformation of enterprises registered in China’s resource-based cities? Further analysis of the mechanism and heterogeneity [68]. The measurement dimensions of environmental performance (efflnincom) are expanded into four levels: public welfare, work rewards and penalties, information disclosure, and pollution control. The specific scores of eight detailed variables are used to describe the environmental performance of companies. Data related to environmental performance from the CSMAR database are used as proxy variables, and the corporate green transformation index is obtained by summing these scores.

3.2.3. Board Structure

This paper measures corporate governance structure from three aspects. As mentioned earlier, companies with a higher proportion of independent directors, a higher proportion of female directors, and a lower shareholding ratio of the largest shareholder are considered to have a governance structure with a longer-term vision. Specifically, this paper uses the proportion of independent directors to the total number of directors as the proportion of independent directors (Indrcrat), the proportion of female directors to the total number of directors as the proportion of female directors (Feldrcrat), and the shareholding ratio of the largest shareholder to the total number of shares as the largest shareholder’s ownership ratio (LrgHldRt).

3.2.4. Capital Market Attention

This paper uses analyst attention to companies as a proxy variable for capital market attention. The number of research reports published by securities analysts on a company is used as a measure to quantify analyst attention (gaze).

3.3. Control Variables

Referring to existing research, the following control variables are selected: firm size (natural logarithm of total assets), cash ratio (CR: ending balance of cash and cash equivalents/current liabilities), tangible asset ratio (TAR: total tangible assets/total assets), intangible asset ratio (RIA: net intangible assets/total assets), fixed asset ratio (FAR: net fixed assets/total assets), current asset ratio (CAR: total current assets/total assets), working capital ratio (WCR: (current assets − current liabilities)/total assets), inventory turnover ratio (RST: cost of sales/average net inventory balance), fixed asset turnover ratio (TFR: revenue/ending balance of net fixed assets), and effective tax rate (ERT: income tax expense/pre-tax profit) [69].

3.4. Construction of the Econometric Model

Double/Debiased Machine Learning (DML), proposed by [70], is a method that improves the estimation of the non-parametric parts of models such as Partially Linear Regression (PLR) by leveraging the predictive advantages of machine learning in high-dimensional scenarios [71]. In this paper, DML not only effectively addresses the endogeneity problem caused by confounding factors in causal inference by analyzing the variables that influence policy distortions but also eliminates regularization bias and overfitting bias when estimating treatment effects using complex machine-learning methods in high-dimensional environments through orthogonalization and cross-fitting. As a result, consistent estimates of the effect of ESG rating divergence on corporate green transformation are obtained [72].

This paper first constructs a simple, partially linear machine-learning model as follows:

$$Y_{i,t}={\theta }_0{ESG\_dif}_{i,t}+{g(X}_{i,t})+U_{i,t}$$

(1)

$$E\left(U_{i,t}|{Policy}_{i,t},X_{i,t}\right)=0$$

(2)

Let i represent the firm, and t represent the year. $Y_{i,t}$ denotes the dependent variable, which is the efficiency of green transformation in energy companies; ${ESG\_dif}_{i,t}$ is the policy variable, and ${\theta }_0$ is the coefficient of interest. $X_{i,t}$ represents the set of high-dimensional variables to be estimated, specifically comprising a collection of control variables. $U_{i,t}$ is the error term with a conditional mean of zero. We estimate Equations (1) and (2), resulting in the following parameter estimates:

$$\widehat{\theta }={\left({\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{{ESG\_dif}_{i,t}}^2\right)}^{-1}{\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{ESG\_dif}_{i,t}(Y_{i,t}-{\widehat{g}(X}_{i,t}))$$

(3)

where n is the sample size. In general machine-learning models, after regularization, the given half of the sample information can be used to estimate the other half using the above formula.

However, the above expression suffers from regularization estimation bias. The random effects bias and factor influence bias are as follows:

$$\sqrt{n}\left(\widehat{{\theta }_0}-{\theta }_0\right)={\left({\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{{ESG\_dif}_{i,t}}^2\right)}^{-1}{\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{ESG\_dif}_{i,t}{U}_{i,t}+{\left({\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{{ESG\_dif}_{i,t}}^2\right)}^{-1}{\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{ESG\_dif}_{i,t}\left[{g(X}_{i,t}\right)-{\widehat{g}(X}_{i,t}\left)\right]$$

(4)

It should be noted that a = ${\left({\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{{ESG\_dif}_{i,t}}^2\right)}^{-1}{\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{ESG\_dif}_{i,t}{U}_{i,t}$ represents the bias under the influence of random factors, and it follows a Gaussian distribution.

On the other hand, b = ${\left({\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{{ESG\_dif}_{i,t}}^2\right)}^{-1}{\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{ESG\_dif}_{i,t}\left[{g(X}_{i,t}\right)-{\widehat{g}(X}_{i,t}\left)\right]$ represents the bias that exists after regularization, which is typically divergent [70]. To obtain a convergent unbiased estimate of the parameter, we employ cross-fitting estimation methods.

We construct a set of regressions for auxiliary cross-fitting:

$${ESG\_dif}_{i,t}={m(X}_{i,t})+V_{i,t}$$

(5)

$$E\left(V_{i,t}|X_{i,t}\right)=0$$

(6)

Here, m denotes the coefficients of the high-dimensional parameters to be estimated, and we express its estimate as ${\widehat{m}(X}_{i,t})$. $V_{i,t}$ also represents a random error term, following a normal distribution with a mean of zero.

We use the first machine-learning process to fit m, yielding the residual term: ${\widehat{V}}_{i,t}={ESG\_dif}_{i,t}-{\widehat{m}(X}_{i,t})$; In this regression process, ${\widehat{V}}_{i,t}$ reflects the influence of $X_{i,t}$ deviating from ${Policy}_{i,t}$.

Next, we utilize the second machine-learning process to fit the final parameter to be estimated g:

$$\widehat{\theta }={\left({\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{{V_{i,t}ESG\_dif}_{i,t}}^2\right)}^{-1}{\displaystyle \frac{1}{n}}{\sum }_{i\in I,t\in T}{V}_{i,t}(Y_{i,t}-{\widehat{g}(X}_{i,t}))$$

(7)

Similar to Equation (4), we can approximate Equation (7) as:

$$\sqrt{n}\left(\widehat{{\theta }_0}-{\theta }_0\right)=[E{\left({V_{i,t}}^2\right)]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\sum }_{i\in I,t\in T}{V}_{i,t}{U}_{i,t}+[E{\left({V_{i,t}}^2\right)]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\sum }_{i\in I,t\in T}\left[{m(X}_{i,t}\right)-{\widehat{m}(X}_{i,t}\left)\right]\left[{g(X}_{i,t}\right)-{\widehat{g}(X}_{i,t}\left)\right]$$

(8)

In this context, $[E{\left({V_{i,t}}^2\right)]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\sum }_{i\in I,t\in T}{V}_{i,t}{U}_{i,t}$ follows a normal distribution with a mean of zero. After two rounds of estimation, $\left[E{\left({V_{i,t}}^2\right)]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\sum }_{i\in I,t\in T}\left[{m(X}_{i,t}\right)-{\widehat{m}(X}_{i,t}\left)\right]\right[{g(X}_{i,t})$ does not exhibit a slowing convergence rate, allowing us to obtain a consistent unbiased estimate.

4. Empirical Analysis

4.1. Analysis of Main Effects and Moderating Effects

In Section 4.1, the conclusions derived from

Table 1. Main Effects and Moderating Effects.

	(1) Z1LM	(2) Z1ML	(3) T1LM	(4) T1ML	(5) T2LM	(6) T2ML	(7) T3LM	(8) T3ML	(9) T4LM	(10) T4ML
(Intercept)	0.359 **		0.384 **		−0.337 **		−0.351 **		−0.340 **
	(0.113)		(0.119)		(0.114)		(0.118)		(0.113)
ESGdif6	0.015 *	0.021 ***	0.017 ***	0.001 **	0.027 ***	0.001 ***	0.019 ***	0.002 ***	0.016 ***	0.011 ***
	(0.006)	(0.001)	(0.002)	(0)	(0.001)	(0.000)	(0.005)	(0.000)	(0.002)	(0.001)
size	0.333 ***		0.333 ***		0.331 ***		0.332 ***		0.328 ***
	(0.006)		(0.006)		(0.007)		(0.007)		(0.007)
CR	−0.158 ***		−0.155 ***		−0.155 ***		−0.156 ***		−0.146 ***
	(0.044)		(0.044)		(0.044)		(0.045)		(0.044)
TAR	−0.013 *		−0.013		−0.012		−0.013 *		−0.011
	(0.007)		(0.007)		(0.007)		(0.007)		(0.007)
RIA	−0.031		−0.030		−0.042		−0.037		−0.010
	(0.117)		(0.117)		(0.118)		(0.119)		(0.117)
FAR	0.102		0.100		0.083		0.096		0.112
	(0.121)		(0.121)		(0.122)		(0.123)		(0.121)
CAR	0.196 ***		0.199 ***		0.199 ***		0.195 ***		0.182 ***
	(0.038)		(0.038)		(0.038)		(0.038)		(0.039)
WCR	0.301 ***		0.298 ***		0.299 ***		0.299 ***		0.299 ***
	(0.050)		(0.051)		(0.051)		(0.051)		(0.050)
RST	−0.177 ***		−0.176 ***		−0.178 ***		−0.175 ***		−0.186 ***
	(0.048)		(0.048)		(0.048)		(0.049)		(0.048)
TFR	0.010 ***		0.010 ***		0.010 ***		0.010 ***		0.010 ***
	(0.001)		(0.001)		(0.001)		(0.001)		(0.001)
ERT	0.000		0.000		0.000		0.000		0.000
	(0.000)		(0.000)		(0.000)		(0.000)		(0.000)
Etaxrt	−0.000		−0.000		−0.000		−0.000		0.000
	(0.003)		(0.003)		(0.003)		(0.003)		(0.003)
Concurrent Position	−0.010		−0.010		−0.009		−0.010		−0.009
	(0.011)		(0.011)		(0.011)		(0.011)		(0.011)
Indrcrat2			0.059 ***
			(0.009)
ESGdif6:Indrcrat2			0.067 ***
			(0.006)
Feldrcrat					0.003 ***
					(0.001)
ESGdif6:Feldrcrat					0.027 ***
					(0.000)
LrgHldRt							0.009 ***
							(0.000)
ESGdif6:LrgHldRt							0.0227 ***
							(0.000)
gaze									0.001 ***
									(0.000)
ESGdif6:gaze									0.039 ***
									(0.001)
R2	0.903		0.903		0.904		0.903		0.905
Adj. R2	0.901		0.900		0.901		0.900		0.901
Num. obs.	1081		1081		1081		1081		1081

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. * p < 0.05. ** p < 0.01. *** p < 0.001.

and

Table 2. Mechanism Test of Green Innovation.

	(1) Efflnincom	(2) Greinn	(3) Efflnincom
(Intercept)	3.168 ***	24.216 ***	3.168 ***
	(0.003)	(4.249)	(0.003)
size	0.333 ***	86.750 ***	0.321 ***
	(0.006)	(7.964)	(0.007)
CR	−0.158 ***	−208.631 ***	−0.130 **
	(0.044)	(54.126)	(0.044)
TAR	−0.013 *	−10.235	−0.012
	(0.007)	(8.297)	(0.007)
RIA	−0.031	−189.589	−0.005
	(0.117)	(145.128)	(0.116)
FAR	0.102	−354.431 *	0.149
	(0.121)	(149.670)	(0.120)
CAR	0.196 ***	−32.628	0.201 ***
	(0.038)	(47.232)	(0.038)
WCR	0.301 ***	47.771	0.295 ***
	(0.050)	(62.407)	(0.050)
RST	−0.177 ***	−98.288	−0.164 ***
	(0.048)	(59.335)	(0.047)
TFR	0.010 ***	1.351	0.009 ***
	(0.001)	(1.092)	(0.001)
ERT	0.000	−0.088	0.000
	(0.000)	(0.343)	(0.000)
Etaxrt	−0.000	0.443	−0.000
	(0.003)	(3.463)	(0.003)
ConcurrentPosition	−0.010	26.841 *	−0.014
	(0.011)	(13.219)	(0.011)
ESGdif6	0.015 *	4.662 ***	0.015 *
	(0.006)	(0.483)	(0.006)
greinn			0.002 ***
			(0.000)
R2	0.903	0.256	0.906
Adj. R2	0.901	0.236	0.903
Num. obs.	1081	1081	1081

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. * p < 0.05. ** p < 0.01. *** p < 0.001.

are discussed separately. Table 1 presents the descriptive statistics and correlation analysis, which provide a foundational understanding of the data characteristics and relationships between key variables. The discussion here focuses on the general trends observed from the data in Table 1, offering insights into the underlying patterns.

Following that, Table 2 presents the results from the regression analysis, which tests the hypotheses and provides deeper insights into the relationships between ESG rating discrepancies and the green transformation efficiency of energy enterprises. It is important to note that the conclusions drawn from Table 1 are based on descriptive and correlational analyses, whereas the conclusions from Table 2 are specifically tied to the regression models and statistical tests.

Columns 1 and 2 of Table 1 present the impacts of ESG rating discrepancies on firms’ green transformation using linear regression and double machine-learning algorithms, respectively. Columns 3 and 4 display the results of the moderating effects of independent director ratios on the main effects obtained from the two methods. Columns 5 and 6 show the moderating effects of board gender ratios on the main effects derived from both approaches. Columns 7 and 8 illustrate the moderating effects of ownership equilibrium on the main effects using the two methodologies. Finally, columns 9 and 10 present the moderating effects of analyst attention on the main effects as determined by both methods.

From Table 2, it can be observed that ESG rating discrepancies positively influence the efficiency of green transformation in energy companies, validating Hypothesis 1. Specifically, a one-unit increase in ESG rating discrepancy corresponds to an average 15.8% improvement in green transformation efficiency, as demonstrated by the results in Table 2, Column 2. Additionally, in energy companies with more balanced governance structures, such as those with higher proportions of independent directors, this effect is magnified to a 21.3% improvement. Additionally, each moderating variable positively enhances the main effects, indicating that internal governance within firms can increase their responsiveness to market supervision, thereby improving substantial environmental protection measures undertaken by the firms.

4.2. Robustness Checks

This study conducted a series of robustness checks to validate the consistency of our findings. Due to space constraints, these results are not presented in the main text but are shown in Appendix A. First, we replaced the neural network algorithm with a random forest model, as specified in the DMLLZU package 1.0.0 on CRAN, and the results, presented in Appendix A 

Table A1. Random forest algorithm results.

	(1) Z1ML	(2) T1ML	(3) T2ML	(4) T3ML	(5) T4ML
ESGdif	0.013 ***
	(0.001)
ESGdif6:Indrcrat2		0.001 ***
		(0)
ESGdif6:Feldrcrat			0.002 ***
			(0)
ESGdif6:LrgHldRt				0.001 ***
				(0)
ESGdif6:gaze					0.001 ***
					(0)

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. *** p < 0.001.

, indicated that the conclusions remained unchanged in terms of significance, though the coefficient magnitudes differed. Additionally, we employed a stacking bagging model, as shown in Appendix A 

Table A2. Bagging model algorithm results.

	(1) Z1ML	(2) T1ML	(3) T2ML	(4) T3ML	(5) T4ML
ESGdif	0.116 ***
	(0.023)
ESGdif6:Indrcrat2		0.003 ***
		(0.001)
ESGdif6:Feldrcrat			0.002 *
			(0.001)
ESGdif6:LrgHldRt				0.002 ***
				(0)
ESGdif6:gaze					0.005 ***
					(0.001)

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. * p < 0.05. *** p < 0.001.

, which confirmed the robustness of the results.

To address potential omitted variable bias, we included two control variables: economic policy uncertainty and the number of employees [73]. Economic policy uncertainty affects corporate decision-making, particularly under economic pressure, influencing the efficiency of green transformation [74,75]. The inclusion of these control variables did not alter the original conclusions, as shown in Appendix A 

Table A3. Adjust the variable quantization method to change the results.

	(1) Z1ML	(2) T1ML	(3) T2ML	(4) T3ML	(5) T4ML
ESGdif	0.005 ***
	(0.001)
ESGdif6:Indrcrat2		0.001 ***
		(0.000)
ESGdif6:Feldrcrat			0.001 ***
			(0)
ESGdif6:LrgHldRt				0.001 ***
				(0)
ESGdif6:gaze					0.003 ***
					(0.001)

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. *** p < 0.001.

.

Furthermore, we modified the quantification method for green transformation efficiency by replacing total revenue with sales revenue in the calculations. The findings, presented in Appendix A 

Table A4. The quantization method of dependent variable changes the results.

	(1) Z1ML	(2) T1ML	(3) T2ML	(4) T3ML	(5) T4ML
ESGdif	0.013 ***
	(0.001)
ESGdif6:Indrcrat2		0.001 ***
		(0)
ESGdif6:Feldrcrat			0 ***
			(0)
ESGdif6:LrgHldRt				0.002 ***
				(0)
ESGdif6:gaze					0.001 ***
					(0)

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. *** p < 0.001.

, remained consistent with the benchmark regression, further supporting the robustness of the results. Lastly, we adjusted the quantification method for ESG rating discrepancies, as shown in Appendix A 

Table A5. The quantization method of the argument changes the results.

	(1) Z1ML	(2) T1ML	(3) T2ML	(4) T3ML	(5) T4ML
ESGdif	0.019 **
	(0.009)
ESGdif6:Indrcrat2		0.001 ***
		(0)
ESGdif6:Feldrcrat			0 ***
			(0)
ESGdif6:LrgHldRt				0 ***
				(0)
ESGdif6:gaze					0.001 ***
					(0)

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. ** p < 0.01. *** p < 0.001.

, and observed consistent conclusions, confirming the robustness of our analysis.

4.3. Mechanism Analysis

The preceding analysis has thoroughly examined the causal relationship between ESG discrepancies and firms’ green transformation [76]. A key question remains: How do ESG discrepancies influence firms’ green transformation? A review of existing literature reveals that a crucial factor affecting corporate transformation, which requires additional attention, is the role of green innovation. As firms enhance production efficiency and reduce carbon emissions through green innovation, many innovative outcomes can improve existing emission equipment, thereby advancing firms’ green transformation. Since ESG discrepancies motivate firms to cater to ESG rating agencies, this results in green innovation and clean production transformation. Therefore, this paper posits that corporate green innovation can serve as a positive mediating factor influencing the relationship between ESG discrepancies and firms’ green transformation. We utilized the CSMAR database to obtain data on corporate green innovation and constructed a mediation model to analyze its mediating role. As shown in the table below, corporate green innovation plays a positive mediating role in the positive relationship between ESG rating discrepancies and the efficiency of green transformation in energy companies. In other words, corporate green innovation enhances market oversight of firms, thus strengthening the positive impact of ESG discrepancies on the efficiency of green transformation in energy companies to some extent. The following conclusion was also validated through 1000 bootstrap sampling tests, yielding the same results.

4.4. Further Research

In addition to the impact of ESG rating discrepancies on corporate green transformation (see

Table 3. Further Research: Analysis of Factors Affecting Corporate ESG Rating Discrepancies and Stock Price Volatility.

	(1) MeanAmret	(2) Efflnincom	(3) MeanAmret
(Intercept)	1.026 ***	3.168 ***	1.026 ***
	(0.150)	(0.003)	(0.150)
size	−0.384	0.333 ***	−1.686 *
	(0.281)	(0.006)	(0.727)
CR	2.586	−0.158 ***	3.203
	(1.911)	(0.044)	(1.932)
TAR	−0.196	−0.013 *	−0.144
	(0.293)	(0.007)	(0.293)
RIA	7.126	−0.031	7.246
	(5.124)	(0.117)	(5.109)
FAR	12.185 *	0.102	11.787 *
	(5.284)	(0.121)	(5.272)
CAR	2.052	0.196 ***	1.283
	(1.668)	(0.038)	(1.709)
WCR	−0.164	0.301 ***	−1.343
	(2.203)	(0.050)	(2.279)
RST	3.868	−0.177 ***	4.562 *
	(2.095)	(0.048)	(2.119)
TFR	0.012	0.010 ***	−0.025
	(0.039)	(0.001)	(0.043)
ERT	0.000	0.000	−0.001
	(0.012)	(0.000)	(0.012)
Etaxrt	0.089	−0.000	0.090
	(0.122)	(0.003)	(0.122)
ConcurrentPosition	0.288	−0.010	0.327
	(0.467)	(0.011)	(0.466)
ESGdif6	−0.116 ***	0.015 *	−0.106 ***
	(0.064)	(0.006)	(0.065)
efflnincom			3.915 ***
			(1.015)
R2	0.037	0.903	0.045
Adj. R2	0.011	0.901	0.017
Num. obs.	1081	1081	1081

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. * p < 0.05. *** p < 0.001.

), do they also have subsequent effects on corporate performance due to green transformation? Specifically, as shown in

Table 4. Further Research: Analysis of Factors Influencing the Relationship Between Corporate ESG Rating Discrepancies and Growth Potential.

	(1) GRO	(2) Efflnincom	(3) GRO
(Intercept)	0.054 **	3.168 ***	0.054 **
	(0.021)	(0.003)	(0.021)
size	−0.068	0.333 ***	−0.258 *
	(0.039)	(0.006)	(0.101)
F011201A	0.091	−0.158 ***	0.181
	(0.265)	(0.044)	(0.268)
F010401A	−0.049	−0.013 *	−0.041
	(0.041)	(0.007)	(0.041)
F031001A	0.270	−0.031	0.287
	(0.710)	(0.117)	(0.708)
F030901A	1.367	0.102	1.309
	(0.732)	(0.121)	(0.730)
F030801A	−0.293	0.196 ***	−0.404
	(0.231)	(0.038)	(0.237)
F030101A	−0.543	0.301 ***	−0.714 *
	(0.305)	(0.050)	(0.316)
F030501A	0.363	−0.177 ***	0.464
	(0.290)	(0.048)	(0.293)
F041401B	0.010	0.010 ***	0.004
	(0.005)	(0.001)	(0.006)
F040505C	0.001	0.000	0.000
	(0.002)	(0.000)	(0.002)
Etaxrt	−0.002	−0.000	−0.002
	(0.017)	(0.003)	(0.017)
ConcurrentPosition	0.026	−0.010	0.031
	(0.065)	(0.011)	(0.065)
ESGdif6	0.023 ***	0.015 *	0.024 ***
	(0.007)	(0.006)	(0.037)
efflnincom			0.569 *
			(0.279)
R2	0.075	0.903	0.084
Adj. R2	0.050	0.901	0.056
Num. obs.	1081	1081	1081

Note. Unstandardized regression coefficients are displayed, with standard errors in parentheses. * p < 0.05. ** p < 0.01. *** p < 0.001.

, ESG rating discrepancies positively influence corporate growth potential while reducing stock price volatility, suggesting that ESG discrepancies serve as important market oversight signals. To explore this, this paper considers the effects of ESG rating discrepancies on both corporate development uncertainty and corporate development returns. Specifically, we use stock price volatility as a proxy for corporate development uncertainty and growth indicators as a proxy for corporate development returns for analysis. Furthermore, we examine whether corporate green transformation can serve as a mediating variable that allows firms to mitigate the negative impact of market rating discrepancies by enhancing their green transformation, thereby improving their growth potential and reducing uncertainty.

As shown in the table below, ESG rating discrepancies negatively impact corporate stock price volatility, indicating that ESG rating discrepancies serve as an important market oversight signal for firms, reducing stock price uncertainty and increasing the market’s generally cautious outlook on firms. Corporate green transformation acts as a positive mediating factor, suggesting that green transformation enhances business stability, thereby increasing the influence of market perceptions on stock price volatility.

As shown in the table below, ESG rating discrepancies can positively influence corporate growth potential. This paper analyzes this situation, and relevant research suggests that ESG rating discrepancies, to some extent, represent disparities in market attention. Therefore, they symbolize corporate development potential and controversy, with increased ESG discrepancies contributing to enhanced corporate growth. Corporate green transformation serves as a positive mediating factor, indicating that green transformation can increase corporate growth potential by reducing institutional costs faced by firms in the context of environmental regulations and enhancing opportunities for transformational development.

5. Discussion

In the context of current climate change and dual carbon policies, the efficiency of green transformation in energy enterprises has become a crucial indicator of their sustainable development capabilities. For instance, energy companies like those in the oil and gas sector can become advocates for energy transition by embracing climate change and taking proactive measures to address its impacts [77]. Ref. [78] also noted that oil and gas companies must develop strategies to adapt to climate change and renewable energy to ensure sustainability in the global energy transition context [78]. Although this indicator is not commonly included in ESG ratings, it reflects how energy companies respond to these challenges. This paper utilizes panel data from Chinese listed companies to explore how ESG rating discrepancies affect the efficiency of green transformation in energy enterprises and further analyze their profound impact on sustainable development. The study employs benchmark regression and double machine-learning model regression to test hypotheses and strengthen the credibility of the research through robustness checks. This paper not only replaces regression methods and core explanatory variables but also adds control variables such as macroeconomic policy uncertainty and the number of employees in energy enterprises. Additionally, environmental violation cases are used as instrumental variables to address endogeneity issues. Finally, the study examines the impact of ESG rating discrepancies on energy enterprises, including their effect on corporate uncertainty (e.g., stock price volatility) and corporate performance (e.g., financial performance) [79]. The empirical results reveal several key conclusions. First, ESG rating discrepancies significantly enhance the efficiency of green transformation in energy enterprises. Second, this positive impact is more pronounced in energy enterprises with a more balanced board structure. Lastly, energy enterprises with higher market attention can also facilitate this positive effect. These hypotheses provide clear directions for empirical research. Finally, through robustness checks, mechanism analyses, and further investigations into corporate impacts, the study enhances the credibility and interpretability of the results. By replacing regression methods, adding control variables, substituting core explanatory variables, and employing instrumental variable methods to address endogeneity, the robustness of the findings is ensured.

6. Conclusions

In summary, our research is the first to incorporate the comprehensive indicator of green transformation efficiency in energy enterprises, innovatively exploring the moderating role of internal board structure on ESG rating discrepancies [80]. This reveals a new dimension of the impact of ESG rating discrepancies on environmental factors in energy enterprises. This theoretical contribution not only expands the boundaries of research on ESG rating discrepancies and understanding of their effects on environmental factors in energy enterprises but also provides a new assessment perspective for energy companies in their responses to climate change and dual carbon policies. Furthermore, previous studies have rarely considered the impact of internal governance structures on environmental performance in energy enterprises [81,82,83]. This finding offers a new research perspective for optimizing governance structures in energy enterprises and has significant practical implications for their sustainable development. Additionally, the use of double machine-learning model regression allows for more accurate identification of the complex causal relationship between green transformation efficiency and ESG rating discrepancies compared to traditional linear models. This methodological innovation enables a deeper understanding of how comprehensive factors influence the sustainable development of energy enterprises. Moreover, by exploring the mechanism of green innovation, this paper analyzes how ESG rating discrepancies affect the efficiency of green transformation in energy enterprises. Finally, by examining the effects of ESG rating discrepancies on corporate uncertainty and financial performance, it reveals the comprehensive impact on the overall performance of energy enterprises.

At the same time, this research provides new insights into understanding and improving the green transformation efficiency of energy enterprises in ESG ratings. On the one hand, energy enterprises should emphasize green transformation efficiency as a critical indicator of their sustainable development, which not only responds to global climate change and dual carbon policy requirements but also significantly enhances their ESG ratings [84]. By optimizing energy use, energy enterprises can reduce energy consumption, lower operational costs, and improve production efficiency, effectively boosting their ESG scores and enhancing market competitiveness [85]. On the other hand, investors and rating agencies should consider incorporating green transformation efficiency into ESG assessment frameworks to evaluate the sustainable development capabilities of energy enterprises more comprehensively [86]. This can help alleviate the information asymmetry issues arising from ESG rating discrepancies and promote the healthy development of capital markets.

The verification of hypotheses and experimental results in this study also reveals several important management insights. The validation of H1 indicates that ESG rating discrepancies significantly promote the efficiency of green transformation in energy enterprises, emphasizing that under different rating standards for various energy companies, ESG discrepancies can help enterprises monitor themselves more comprehensively. For large-scale energy enterprises with high emissions, discrepancies in ESG ratings can attract broader market attention, thereby motivating these companies to adopt more proactive emission reduction measures and green transformation strategies. Consequently, energy enterprises should pay greater attention to ESG ratings, increasing transparency and enhancing communication with investors to reduce rating discrepancies and gain trust and support from the capital market. The validation of H2 and H3 indicates that a balanced board structure and high capital market attention can enhance this positive effect. Therefore, energy enterprises should strive to optimize their board structures, increasing the proportion of independent directors, achieving gender balance, and reducing the proportion of the largest shareholder to enhance the board’s oversight and decision-making capabilities, enabling more effective responses to ESG rating discrepancies. Finally, through further analysis of the green innovation mechanism, we find that green innovation is an essential mediator linking ESG rating discrepancies and the enhancement of green transformation efficiency in energy enterprises. ESG rating discrepancies can prompt companies to increase R&D investment in clean energy technologies, energy efficiency improvements, and emission reduction technologies, driving innovation in green technologies and products, thus improving green transformation efficiency [87]. Therefore, energy enterprises should regard green innovation as a key strategy for enhancing ESG performance and addressing climate change, accelerating the commercialization of green technologies.

7. Research Limitations

While this study provides valuable insights into the role of ESG rating discrepancies in promoting green transformation efficiency, there are several limitations that should be noted:

(1)

The study focuses on Chinese listed energy companies, which may limit the generalizability of the findings to other regions or industries. Future research could consider cross-country comparisons or other sectors.

(2)

The proxy variable for green transformation efficiency may not capture all dimensions of corporate sustainability performance. Incorporating alternative metrics could enhance robustness.

(3)

Although the use of double machine learning mitigates endogeneity, potential biases from unobserved variables cannot be entirely ruled out.

(4)

The data spans from 2007 to 2022. Future studies could explore more recent data to assess the evolving dynamics of ESG performance.