The Impact of Public Environmental Concern on Corporate ESG Performance

Abstract

Utilizing an advanced machine learning algorithm, particularly the Artificial Neural Network (ANN) framework, this study reveals a significant nonlinear and even cyclical relationship between public concern about environmental issues and the ESG performance of Chinese A-share listed companies, covering the period from 2004 to 2020. The findings highlight the effectiveness of the Self-Organizing Map (SOM)-ANN framework in elucidating the empirical relationship between these variables. We contend that robust public monitoring can enhance companies’ ESG initiatives, and we recommend that policymakers implement a series of measures to safeguard and promote public involvement in decision-making processes. Furthermore, our analysis of the combined effects of public concern and various performance metrics on firms’ ESG outcomes indicates that the diversity among firms is crucial for determining the most appropriate level of public participation in their sustainable development efforts. Therefore, managers and policymakers should focus on firm-specific attributes instead of adopting a “one-size-fits-all” approach to maximize the benefits of public engagement.

1. Introduction

Scholars have highlighted the significance of formal institutions, asserting that they serve as internal catalysts for knowledge accumulation and technological innovation, rather than relying on substantive factors like material and human capital (Acemoglu et al., 2001; Rodrik et al., 2004). More recent studies have concentrated on the underlying informal institutions, yielding initial insights into the influence of public environmental concern (PEC) on company conduct (Gu et al., 2021, 2022). Nonetheless, a microanalysis approach from the perspective of social value is absent, particularly regarding the impact of PEC on corporate environmental, social, and governance (ESG) performance.

Most environmental issues are persistent, extensive, noticeable, and extensively studied (Zhuo et al., 2023; Wu et al., 2023; Qiu et al., 2023; Jin, 2023). The public is now much more concerned about the environment and participates in environmental governance by voicing environmental demands through a variety of channels as environmental protection has become more serious (Yang et al., 2021; Ren et al., 2021; Wu et al., 2024; X. Zhang et al., 2023). As a result, PEC can be thought of as an ordinary informal institution. Businesses are under a lot of pressure in this situation, and they have to modify their development plans to suit the interests of the stakeholders. This study adds to the body of knowledge already available on the connection between PEC and ESG performance. For our research, we used a sample of Chinese-listed firms for a few reasons. First, formal institutions are still in the early stages of China’s economic revolution. Informal institutions are playing a bigger role in social development and business decision-making as useful supplements to pertinent policies (Allen et al., 2005). Second, PEC has been rising yearly as a result of the Chinese government’s deliberate efforts to encourage public participation in environmental governance through the implementation of laws and regulations (Hao et al., 2023). A growing force in social governance is the public’s ability to negotiate more successfully with businesses and the government. Furthermore, PEC varies by region in China due to disparities in geographic location, economic growth, and educational attainment (Liu & Mu, 2016; Ren et al., 2022), all of which are hard to alter quickly (Guiso et al., 2004). Consequently, it is important to take into account the fundamental connection between PEC and Chinese firms’ ESG performance.

This research makes substantial contributions to three primary areas: methodology, theory, and empirical findings. The methodological integration of Artificial Neural Network (ANN) modeling with bootstrapping techniques examines the relationship between firms’ ESG performance and PEC, based on actual variable relationships, thereby obviating the necessity for functional assumptions and addressing endogeneity concerns. This study incorporates variables of current high interest in pertinent research topics to enhance the robustness of the results. The analysis empirically identifies the inflection point of the nonlinear relationship between PEC and corporate ESG performance by scrutinizing data from 862 publicly listed small- and medium-sized firms (SMEs) over 17 years. The discovery offers precise theoretical guidance and empirical validation for politicians, corporate regulators, and investors.

This work is organized as follows: Section 2 offers a theoretical analysis and literature review. The methodology and the data used for the research are thoroughly analyzed in Section 3. Section 4 presents the empirical findings. Finally, Section 5 outlines the conclusions and policy implications.

2. Theoretical Analysis and Literature Review

2.1. Theoretical Background

Our research is based on two independent theoretical frameworks: (1) organizational behavior and decision-making, and (2) real options theory. Agency theory provides a robust framework for understanding how PEC affects strategic changes in ESG activities. In the context of increased PEC, organizations may face intensified agency conflicts between managers and shareholders (El Ghoul et al., 2024). This tension arises from the conflicting interests of the two groups, since managers may emphasize their personal preferences or short-term objectives over the long-term interests of shareholders. Agency theory suggests that managers may intentionally employ ESG performance metrics to mitigate this discrepancy and reduce agency conflicts. By investing in ESG performance that promotes social and environmental goals, managers can exhibit their commitment to long-term value creation and stakeholder interests, thereby aligning their aims with those of shareholders and reducing agency costs (Zhu & Wagner, 2024). These signaling approaches enhance transparency, foster trust, and bolster businesses’ reputational capital, thus increasing shareholder confidence and mitigating the risk of opportunistic managerial conduct.

Likewise, stakeholder theory offers an additional explanation that PEC may influence firms’ ESG initiatives by shaping stakeholder expectations and pressures. Figueira et al. (2023) suggest that stakeholders may reassess their goals, emphasizing social and environmental considerations, and want corporations to demonstrate increased commitment to these issues. By aligning their ESG initiatives with stakeholder expectations and societal demands, organizations can cultivate positive relationships with key stakeholders, enhancing trust and goodwill.

Institutional theory provides a robust framework for examining the interplay between PEC and corporate responses to ESG activities, considering the impact of institutional pressures from many stakeholders, including regulators, industry peers, and civil society organizations. Institutional forces profoundly influence firms’ ESG initiatives, requiring compliance with established standards and expectations for ethical corporate conduct (Kirste et al., 2024). Companies may face reputational risks and stakeholder backlash if they fail to adhere to societal norms or address pressing social and environmental challenges. The institutional theory asserts that PEC intensifies institutional forces, compelling organizations to prioritize ESG performance as strategic actions to mitigate risks, enhance legitimacy, and navigate volatile socio-economic contexts.

2.2. Corporate ESG Performance

Corporate governance, social responsibility, and environmental impact are the three areas in which the sustainable growth of businesses is given priority in the innovative evaluation framework known as the ESG performance of organizations. Although ESG research has only recently begun in China, it has received a lot of momentum and substantial attention from the government and other societal sectors in recent years. ESG rating and ESG investing are the two main topics of the current corpus of research in the ESG field. Multiple studies have been conducted using ESG data provided by stock exchanges. These studies have analyzed several factors, including the impact of ESG expenditures on corporate financial performance (Wang & Sarkis, 2017), the costs related to financing (Chang et al., 2021; Daryaei et al., 2024), and the possibility of innovation (Dong et al., 2022). Multiple studies have been carried out to authenticate the economic benefits of ESG investments from different perspectives (Broadstock et al., 2021; Duque-Grisales & Aguilera-Caracuel, 2021; Feng et al., 2022; Li et al., 2018; Ren et al., 2023a). Nevertheless, there has been insufficient focus on the determinants that impact business ESG performance.

The creation of rating models and the establishment of ESG assessment standards are the main areas of interest for ESG rating research. The mainstream rating approach, which emphasizes the environment, society, and governance, is an extensive evaluation of a business’s non-financial risks and sustainable development. It is a useful tool that is consistent with the idea of high-quality development and may be used to assess company ESG performance. Currently, a number of agencies are involved in the evaluation of business ESG ratings. However, due to a variety of scoring standards, industry modifications, and data sources, there is a considerable amount of fluctuation in these ratings between organizations. As a result, there is no correlation between the ESG ratings given by various rating agencies (Chatterji et al., 2016). The results of an ESG rating have a big impact on investors’ choices. Avramov et al. (2022) demonstrate that a company with a higher ESG rating necessitates a lower return from investors. However, the divergent ESG rating results from various rating agencies will partially counterbalance the return requirements.

Regarding studies on ESG investments, Renneboog et al. (2011) point out that socially responsible investment (SRI) funds do not perform better financially, and some researchers contend that even though ESG can reduce systemic risk, ESG investments do not produce higher returns. Riedl and Smeets (2017) provide a study that takes into account investors’ innate social values and the significance of spreading charitable messages. They stress that what promotes sustained ESG performance and improves the company’s public image is the adoption of ESG considerations into these investors’ ownership policies and practices.

2.3. Public Environmental Concern

PEC is identified as a “bottom-up” constraint, emerging from the escalating public attention to environmental issues and the resultant expansion of informal institutions that support sustainability (Langpap & Shimshack, 2010). This constraint impacts government, business, and society, influencing managerial decisions in the business sector. According to reputation theory, managers seek to enhance their social reputation and achieve sustainable operations by reducing emissions (Kathuria, 2007), increasing green investments (Liao & Shi, 2018), and adjusting environmental marketing tactics (Banerjee et al., 2003) to meet public environmental expectations. Furthermore, environmental responsibility is vital for businesses to reconcile relationships with diverse stakeholders. Concurrently, the rising environmental concerns among consumers are driving a shift in corporate production, as companies increasingly offer environmentally friendly goods and services, such as new energy sources (Long et al., 2022).

Secondly, according to H. Zhang et al. (2022), public concerns have the potential to mitigate information asymmetry between the government and the public and enhance the efficacy of regulations. Citizens can increase government attention to environmental issues by “voting with their feet” during migration and “with their hands” during elections, even when informal institutions are not required (Hårsman & Quigley, 2010). The central and local governments are urged to enhance their environmental governance policies through pressure transmission (List & Sturm, 2006) by boosting funding for environmental governance, streamlining industrial structures, and enhancing environmental petition processes.

Lastly, institutional theory makes the case that corporate strategies and behaviors are influenced by norms and perceptions (DiMaggio & Powell, 1983; Scott, 2013). PEC, as a societal norm, may have an impact on how laws are developed and implemented within an industry, which may result in related organizational behavior changes and strategic reactions (H. Zhang et al., 2022). While neglecting the implications of PEC on ESG, the majority of earlier studies focused on how PEC influences corporate and governmental efforts to safeguard the environment. In a fair manner, our research adds to the body of knowledge already available on the effects of PEC. It also offers factual support for the idea that informal institutions affect business practices.

2.4. Public Environmental Concern and Corporate ESG Performance

When the government does not provide much attention, the involvement of the public can make up for the weaknesses of official organizations and, in the end, improve the long-term viability of creating business value (Beierle, 2010). Externally, the heightened public environmental consciousness reflects apprehensions surrounding livelihood matters. This can foster effective communication between individuals and the government, prompting the government to implement more efficient environmental rules to direct firms to embrace environmental social responsibility. Corporations prioritize ESG as they strive to improve their environmental performance. This is because they recognize the need to align their business plans with consistent conduct. Additionally, corporations integrate the enhancement of their corporate image into their development strategies, as highlighted by Porter and Kramer in 2006. In addition, PEC serves as a significant means of regulation, reflecting the restrictions of social legitimacy (List & Sturm, 2006). It also attracts the attention of the media, analysts, and other market participants, thus subjecting corporations to various regulatory pressures (Davis & Marquis, 2005). In order to prevent unfavorable publicity, firms modify practices that have a detrimental impact on stakeholders (Aksak et al., 2016; Carmichael & Brulle, 2017). Additionally, the influence of public morality as a normative pressure further strengthens their recognition of the importance of proactive ESG (Brammer et al., 2012). PEC internally guides company management towards enhancing the development of self-regulatory structures and reducing opportunistic behavior (Khan et al., 2013). Simultaneously, it provides guidance to the company’s management in formulating conscientious plans and groundbreaking concepts, motivating them to actively embrace ESG (Shrivastava, 1995; Li, 2022). Lin et al. (2024) reveal that ESG implementation can significantly promote corporate green technology innovation, and public environmental concern will enhance the promotion effect. Zhao et al. (2023) find that the greater the online search attention given to firms, the better the ESG is of the managers, especially for the firms involved in negative news. Ren and Ren (2024) examine the impact of public environmental concern on corporate ESG performance. The findings indicate that heightened public awareness of environmental issues significantly improves the level of corporate ESG performance. Chen et al. (2024) use a textual analysis approach to investigate the impact of public climate concern on corporate ESG disclosure. The results reveal a significant negative correlation between the two. Tao et al. (2023) studied the effects of informal institutions of PEC on corporate social responsibility (CSR) regarding public environmental concern and the results suggest that PEC significantly and positively impacts CSR.

These findings underscore the complexity of the PEC-ESG interaction and the need for a more nuanced understanding of the factors affecting companies’ responses to PEC. Despite the valuable insights provided by contemporary research, the potential nonlinear dynamics between PEC and corporate ESG performance remain largely unexplored. An empirical investigation of this gap, coupled with an analysis of the boundary conditions and mechanisms that regulate the nonlinear relationship between PEC and ESG performance, can augment future research’s comprehension of how firms navigate the challenges and opportunities presented by PEC in their pursuit of sustainable and responsible business practices.

To address this gap, we propose the following research hypothesis:

Hypothesis 1.

There is a nonlinear relationship between PEC and corporate ESG performance.

3. Data and Methodology

3.1. Data

SMEs are indispensable to the green and low-carbon transformation of the economy and inclusive development, given the large number of SMEs (31.2% of Chinese A-share listed companies) in the national economy and the large number of people they involve in terms of employment. Also, the ESG management of large corporations cannot be separated from the SMEs in their value chains. This paper focuses on the influence of PEC on the ESG performance of SMEs in China, using data from 2004 to 2020. This study conducts a screening process that involves the following steps: (i) This excludes enterprises that are listed in the financial sector, and (ii) this also excludes firms that have abnormal trading status, such as ST, *ST, PT, etc. Following the filtering process, the sample consists of 862 firms, with a total of 14,654 annual observations of these firms being retained. We strive to incorporate a wide range of independent variables in our analysis due to the absence of multicollinearity concerns in Artificial Neural Networks (ANNs). The research includes a list of independent variables, which can be found in

Table 1. Independent variables.

No.	Variables	Definitions	Similar Data Employed in Previous Studies
1	Public environmental concern	Average daily search volume of the Baidu Index	(Zhao et al., 2023; Lin et al., 2024; Ren & Ren, 2024; Chen et al., 2024; Tao et al., 2023)
2	Firm size	Logarithm of a firm’s total assets	(Chen et al., 2022; Huang et al., 2023; X. Zhang et al., 2023; Mu et al., 2023; Ren et al., 2023a; Chang et al., 2021; Broadstock et al., 2021; Drempetic et al., 2020; Feng et al., 2022; Houston & Shan, 2022; Huang et al., 2022; Jiang et al., 2022; Ji et al., 2022; Nirino et al., 2021; Wong et al., 2021; Tian et al., 2022; Xia et al., 2022; Wu & Huang, 2022; Dai & Zhang, 2023)
3	ROA	Ratio of a firm’s net profit to total assets	(Chen et al., 2022; Huang et al., 2023; Mu et al., 2023; Ren et al., 2023b; Chang et al., 2021; Feng et al., 2022; Houston & Shan, 2022; Huang et al., 2022; Ji et al., 2022; Nirino et al., 2021; Wong et al., 2021; Xia et al., 2022; Wang & Sarkis, 2017)
4	Firm age	Logarithm of firm age	(X. Zhang et al., 2023; Mu et al., 2023; Chang et al., 2021; Ji et al., 2022; Wu & Huang, 2022)
5	Leverage ratio	Ratio of total liabilities to total assets	(Huang et al., 2023; X. Zhang et al., 2023; Mu et al., 2023; Ren et al., 2023b; Chang et al., 2021; Broadstock et al., 2021; Drempetic et al., 2020; Feng et al., 2022; Houston & Shan, 2022; Huang et al., 2022; Jiang et al., 2022; Ji et al., 2022; Nirino et al., 2021; Li et al., 2018; Wong et al., 2021; Tian et al., 2022; Xia et al., 2022; Wang & Sarkis, 2017; Dai & Zhang, 2023)
6	ROE	Ratio of net income to shareholders’ equity	(Drempetic et al., 2020; Nirino et al., 2021; Tian et al., 2022; Dai & Zhang, 2023)
7	Tobin’s Q	Ratio of a firm’s market value to its replacement cost	(Huang et al., 2023; Dai & Zhang, 2023; Ren et al., 2023b; Houston & Shan, 2022; Huang et al., 2022; Nirino et al., 2021; Li et al., 2018; Wong et al., 2021; Tian et al., 2022; Wang & Sarkis, 2017)
8	NPM	Ratio of net profit to gross revenue	(Drempetic et al., 2020)
9	SOE	Stated owned enterprise (dummy variable)	(X. Zhang et al., 2023; Ren et al., 2023a; Xia et al., 2022; Wu & Huang, 2022)
10	Proportion of independent directors	Ratio of independent directors to directors	(Huang et al., 2023; Mu et al., 2023; Ji et al., 2022; Dai & Zhang, 2023)
11	Fixed asset ratio	Ratio of net fixed assets to long-term funds	(Huang et al., 2023)

. The dependent variable is the corporate ESG performance data obtained from the Hua Zheng ESG rating. The Hua Zheng ESG rating system is based on the core ESG connotation and development experience. It combines the actual situation of the domestic market to build a three-tier index system from top to bottom. Hua Zheng integrates traditional and alternative data to build an algorithm-driven big data engine, which regularly measures the ESG level of listed companies and debt issuers on a quarterly basis and assigns them a nine-grade rating of “AAA-C” accordingly. Hua Zheng’s ESG ratings cover all A-shares and date back to 2009, making it the longest traceable ESG rating system in the Chinese market. The China Stock Market & Accounting Research Database (CSMAR) provides corporate characteristic data.

This work utilizes the research conducted by Zheng et al. (2012) to investigate the topic of “environmental pollution” in all 31 provinces of China. By utilizing the Baidu Index online platform, the daily average search volume for “environmental pollution” by internet users from 2011 to 2020 is obtained. The values are summed and incremented by one, and then the natural logarithm is applied to obtain the data for PEC.

3.2. Methodology

3.2.1. Artificial Neural Network Model

In his groundbreaking work, Hornik (1991) pioneered the application of multilayer feedforward networks, a specific variant of Artificial Neural Networks (ANNs), as universal approximators proficient in replicating the essential functional structure and intrinsic relationships between dependent and independent variables. This foundational research laid the groundwork for a range of technological advancements in the field of artificial intelligence, significantly contributing to the proliferation of applications such as language models (LLMs) exemplified by ChatGPT. These sophisticated models also have demonstrated indispensable utility across various domains including chatbots, autonomous vehicles, machine translation, facial and voice recognition, and other areas reliant on precise predictive capabilities.

In comparison to traditional econometric models, ANNs consistently exhibit superior accuracy as evidenced by their lower Mean Squared Error (MSE). Traditional econometric models often face challenges when dealing with nonlinear functions like trigonometric operations and logarithmic summation of independent variables. Conversely, ANNs excel in emulating diverse functional structures, establishing them as universal approximators renowned for their ability to capture intricate nonlinear relationships among independent variables. Given that many econometric methodologies are based on linear frameworks, ANNs emerge as powerful tools for researchers aiming to explore complex nonlinear associations, as they are proficient in faithfully reproducing underlying functional structures. Moreover, the capacity of ANN models to transcend linear assumptions allows data to autonomously reveal insights without the constraints imposed by predefined functional form assumptions.

Despite the potential applications of machine learning (ML) technologies, particularly Artificial Neural Networks (ANNs), their integration remains limited in economic and financial research (for a detailed discussion on potential applications, see Mullainathan and Spiess (2017)). One plausible explanation for this limitation lies in the challenge of interpreting outcomes for economists well versed in econometric methodologies but less familiar with the intricacies of the ANN paradigm. Addressing this challenge, Cheong et al. (2022) pioneered the Regression by ANN with Bootstrapping (RAB) framework to enhance the dissemination of research findings. By bridging the gap between ANN techniques and traditional econometric models, the RAB methodology aims to streamline the exploration of complex nonlinear relationships among variables. Consequently, both the ANN model and the RAB approach were employed in this study to uncover intricate nonlinear relationships among the variables.

In this study, the ANN model was structured with a dual-stage architecture, comprising a classification neural network in the initial phase and a function approximation mechanism in the subsequent stage, enhancing the model’s accuracy. Initially, a classifier was utilized to partition the data into four distinct groups, ensuring that data points within each group shared similar inherent characteristics. To achieve this segmentation, a Self-Organizing Map (SOM) neural network was employed in the primary phase of this study. The categorized data from the initial stage then proceeded to the subsequent phase, where four separate function approximation ANNs were employed. This bifurcated approach enhanced the efficiency of the training process and improved the accuracy of the function approximation ANN for each group, given the inherent similarities among data points within each subgroup. This design approach has also garnered attention from other researchers, including B. Zhang et al. (2004), Weng et al. (2009), Nourani et al. (2014), Lin et al. (2016), and Cheong et al. (2022).

The combined approach of the Self-Organizing Map (SOM) and Artificial Neural Network (ANN) methods offers a significant advantage over traditional statistical methods by effectively addressing the challenges of data complexity and nonlinearity. The SOM, as an unsupervised learning technique, is first employed to partition the dataset into distinct clusters based on inherent patterns and structures, thereby simplifying dataset handling and ensuring that the forecasting model is applied to more homogeneous subsets of data, which improves its accuracy and relevance. Subsequently, ANN is utilized for forecasting, leveraging its ability to model complex, nonlinear relationships within each cluster. Unlike traditional statistical methods, which often rely on linear assumptions and struggle with high-dimensional or heterogeneous data, the SOM-ANN approach capitalizes on the complementary strengths of both techniques: the ability of the SOM technique to organize and classify data, and the capacity of ANNs for precise, adaptive prediction. This hybrid methodology not only enhances forecasting performance but also provides deeper insights into the underlying data structure, making it a superior alternative to conventional statistical approaches, particularly in scenarios involving complex, nonlinear, and high-dimensional datasets.

The ANN comprises neurons, each governed by a set of formulas. Assuming the neuron receives inputs from i independent variables (X₁, X₂, …, X_i), each neuron is defined by a bias constant and j weights. The initial output of neuron j is determined by the sum of its bias and the weighted sum of its inputs, a relationship that can be mathematically represented by the following equation:

$$S_j=\sum _{k=1}^i{X}_k{W}_{j,k}+B_j$$

(1)

where the weight of input k in neuron j is denoted by W_j,k, while B_j represents the bias within the network. The output of neuron j is then passed through the activation function, also known as the transfer function, which is essentially an equation that maps input to output, to compute the final output.

The choice of activation function often takes the sigmoid form, although it can also appear in various other forms such as logistic, hyperbolic tangent, rectified linear unit, and others. In each of the four function approximation Artificial Neural Networks (ANNs), the inputs are initially directed to the input layer of the ANN architecture. Subsequently, neurons are instantiated in the hidden layer using the aforementioned equations. The outputs from all neurons in the hidden layer are then consolidated in the output layer through an activation function that combines the outputs of the hidden layer neurons. This arrangement, as proved by Hornik (1991), demonstrates the capacity to replicate the functional relationships between dependent and independent variables across various types, thus earning the designation of a “universal approximator.”

To refine the neural network, biases and weights of the neurons are adjusted in each iteration using backpropagation and gradient descent techniques. Backpropagation is a supervised learning algorithm used to train Artificial Neural Networks (ANNs) by efficiently computing the gradient of the loss function with respect to each weight in the network. It works by propagating errors backward from the output layer to the input layer, enabling the network to adjust its weights iteratively and minimize prediction errors. Gradient descent is an optimization algorithm employed to minimize the loss function by iteratively adjusting the model’s parameters in the direction of the steepest descent. It calculates the gradient of the loss function with respect to the parameters and updates them proportionally to the learning rate, ensuring convergence toward the optimal solution. By using backpropagation and gradient descent techniques, errors can be reduced iteratively.

Gradients are calculated by applying the chain rule in conjunction with the backpropagation method. By iteratively updating biases and weights in the direction of the steepest descent, as guided by the negative gradient, the Mean Squared Error (MSE) is minimized towards its local minimum through gradient descent. After each iteration, parameter values are modified to progressively reduce the MSE to the desired level. The gradient descent algorithm can be expressed by the following formula:

$${\theta }_{i,t+1}={\theta }_{i,t}-\alpha {\displaystyle \frac{\partial }{\partial {\theta }_{i,t}}}J\left({\theta }_{i,t}\right)$$

(2)

where ${\theta }_{i,t+1}$ is the value of the parameter ${\theta }_i$ after updating, ${\theta }_{i,t}$ represents the initial value of the parameter ${\theta }_i$ in iteration t before the update, $\alpha$ is the learning rate, $J\left({\theta }_{i,t}\right)$ is the MSE function in terms of ${\theta }_{i,t}$, and ${\displaystyle \frac{\partial }{\partial {\theta }_{i,t}}}J\left({\theta }_{i,t}\right)$ is the gradient of ${\theta }_{i,t}$.

In the domain of machine learning methodologies, an inherent challenge concerns the issue of overfitting, wherein a model excels excessively in replicating the intricacies of the training dataset but struggles when processing entirely new data not included in the training set. As a result, the model’s ability to generalize beyond the training dataset diminishes, rendering it ineffective in accommodating data from diverse sources. Various strategies can be employed to address this challenge, including methods like dropout, early stopping, regularization, and simplifying the model’s architectural complexity. In this study, a constraint was imposed on the neural network’s structural complexity to achieve a balanced trade-off between adaptability and generalization. The subsequent formula outlines the optimal number of neurons for the model:

$$N=({\displaystyle \frac{S{P}_t{N}_O-N_O}{I_O+N_O+1}}-1)/F_L$$

(3)

where S represents the total number of samples available. F_L, as per the practice established by Cheong et al. (2022), serves as a limiting factor to restrict the flexibility of the ANN and is set to a value of 10. N denotes the suggested number of hidden neurons in the ANN, which serves as a reference point for determining the actual number of neurons in the model. P_t represents the proportion of samples utilized during the training process and classified as the training set. Furthermore, N_O represents the number of output neurons, while I_O represents the total number of independent variables.

In the pursuit of objectively evaluating model accuracy, the entire dataset underwent a random split into two separate subsets. The training set comprised 90% of the total data, while the remaining 10% was allocated to the testing set. Following model training on the designated training set, the model’s accuracy was assessed using the testing set. This division of data into distinct subsets is a common practice in machine learning, enabling a fair and unbiased assessment of model effectiveness. The model identified as exhibiting the lowest Mean Squared Error (MSE) when evaluated on the testing set was considered the optimal model.

It is important to note that the issue of endogeneity, which can introduce biased estimates of slope parameters in linear regression analyses, is not relevant to the study of Artificial Neural Networks (ANNs). ANNs function independently of linear relationships, eliminating the need for slope parameters in the model structure. Therefore, the issue of endogeneity, which pertains to slope parameters, does not apply to ANNs. This distinction is crucial as it emphasizes the unique advantages of using ANNs to navigate complex data relationships.

3.2.2. Regression by ANN with Bootstrapping (RAB) Approach

In this study, the RAB (Regression by ANN with Bootstrapping) methodology, as introduced by Cheong et al. (2022), was employed to study the intrinsic relationship between the dependent variable and its determinants. Notably, the RAB technique surpasses traditional econometric analyses by providing deep insights into relationships that are challenging to uncover through standard linear regression methods. The conventional linear regression framework comprises three fundamental components: the Mean Squared Error (MSE) of the linear representation, the slope parameter of the line, and the statistical analysis carried out on this slope parameter. It is evident that the RAB approach outperforms traditional linear regression methodologies across all these dimensions.

A significant difference between the Artificial Neural Network (ANN) model and conventional linear regression lies in their accuracy. Due to its ability to capture complex nonlinear relationships and act as a universal approximator, the ANN model demonstrates significantly improved accuracy compared to the linear regression model. When the underlying relationship is linear, both models produce similar MSE results. However, in cases where nonlinearity is present, the ANN model outperforms the linear regression model in terms of accuracy. Therefore, the linear regression model, limited by its ability to represent only linear relationships, can be seen as the most basic form of an ANN.

In contrast to the standard linear regression model, which calculates the slope parameter (beta) to represent the relationship between the dependent variable and the independent variables, the ANN model does not rely on beta. It is crucial to understand that beta indicates the rate of change in the dependent variable concerning the change in the independent variable, representing the slope parameter. Thus, the traditional linear regression model emphasizes changes in the variables rather than the variables themselves. Another limitation of the linear regression model is its assumption of a consistent slope parameter across the whole range of the independent variables. On the other hand, the RAB methodology employs a two-dimensional figure in presentation which allows for variable slopes across the range, effectively capturing highly nonlinear relationships with high precision. Additionally, in contrast to the focus on changes in variables in the linear regression model, the RAB method enables a comprehensive understanding of variable relationships at the level domain.

The third advantage offered by the RAB methodology is evident in statistical evaluation. While the conventional linear regression model relies on statistical tests centered on the slope parameter, the RAB approach utilizes bootstrapping methods to calculate the confidence interval for the independent variable at the level domain. Specifically, the bootstrapping technique was utilized to generate 6000 samples. Unlike the linear regression model, which emphasizes the confidence interval of the slope parameter over the variable itself, the RAB method directly presents the confidence interval of the independent variable, positioning it as a superior alternative to the conventional linear regression model.

4. Results and Discussion

4.1. The Impact of Public Environmental Concern on Corporate ESG Performance

The relationship between public environmental concern (PEC) and corporate ESG performance as captured by the ESG score is illustrated in Figure 1. As can be observed clearly, quite a strong nonlinear relationship exists between the two variables.

When the value of the PEC index stays below 200, the corporate ESG score initially decreases, followed by a slight increase, with the turning point emerging at around a PEC index value of 100. The ESG score fluctuates around 4.1 at this moment, while it is found to plunge sharply as the value of the PEC index continues to rise. It decreases all the way down to around 3.85 when the PEC index reaches a value of 400, which is also identified as the minimum of the ESG score in the dataset adopted. When the PEC index further increases from 400 to 800, the ESG score keeps rising, though at different speeds. Finally, the score drops once again when the value of the PEC index surpasses the threshold of 800. The maximum value of the ESG score is found to be around 4.4 with a corresponding PEC index of 800.

It is quite thrilling that the relationship between PEC and ESG scores is somewhat cyclical. Periods of contraction alternate with periods of expansion while peaks and troughs occur in turn repeatedly. Additionally, the distance between different confidence intervals is fairly narrow, which indicates a mild level of uncertainty in the model performance.

China has been actively calling for public participation in environmental management in recent decades, and thanks to the popularization of social media, public interest and awareness of environmental protection have been effectively enhanced. Several studies were carried out under this context to examine whether more intense public involvement improves companies’ ESG performance. Nevertheless, most of these studies adopted linear regression techniques by presuming a linear relationship between the two factors. Our work thus contributes to the literature significantly by identifying the nonlinear and even cyclical relationship between PEC and corporate ESG performance.

As an ESG score above 4 is generally considered acceptable for a company’s ESG performance, we identify in Figure 1 where an ESG score would fall below this critical level. It is observed that when the PEC index ranges between 300 to 400 or exceeds 1200, the ESG score remains unpromising, not being able to surpass the benchmark value of 4.

Regarding how an increasing amount of public attention to environmental issues may impact a corporation’s ESG performance, we believe it may share some similarities with parenting styles. It is unlikely for a child to perform well if the child feels suffocated under tight parental supervision. This explains why a firm may perform unsatisfactorily when PEC is too high (over 1200). It indicates that the firm has become overwhelmed by stringent enforcement from the public. At this stage, the firm may feel compelled to prioritize ESG development, which could lead to inefficient resource allocation or even financial distress, further impairing the firm’s ESG performance.

Moreover, for a child who has long suffered from uninvolved parenting, it takes time for the child to get used to suddenly heightened attention from parents. This refers to a “transition period” when PEC is climbing gradually from 300 to 400, the firm may be fragile to overexposure, and it is struggling to find the most suitable solution to increasingly constraining public monitoring.

By referring to Figure 1, one could conclude that a relatively high level of the PEC index, namely, around 800, would be ideal for a firm’s ESG performance. According to the World Bank’s classification of environmental policies, “public participation” currently stands out as the most notable toolkit compared to the other three relatively traditional approaches including “direct regulation”, “market-based measures”, and “market creation”. Given that environmental policies in China have long been executed through direct commands, whose effectiveness is gradually diminishing, it is of vital importance for policymakers to further strengthen the transmission mechanism of environmental policies by integrating them with “public participation”.

Next, we present the following policy suggestions. Firstly, to ensure trustworthiness, laws and regulations must be enacted to protect privacy and encourage public expression, allowing individuals to feel secure when voicing their opinions on various platforms. This may prove challenging for a country like China, which has traditionally adopted “top-down” rather than “bottom-up” policies. In this regard, policymakers might consider the idea of incrementalism originally proposed by the political scientist Charles E. Lindblom. By initiating small policy changes that can lead to more significant shifts over time, this approach ensures a smoother transition during changes in the policy regime, though it may require a longer implementation horizon. The central government may consider gradually enhancing penalties for violations of laws or regulations, providing the public with a buffer period to acclimate to policy adjustments.

Secondly, to improve transparency, current platforms or systems designed for collecting public opinions must be upgraded. The most appealing aspect of “bottom-up” policies is the significant reduction in information asymmetry that leads to more effective external monitoring of firms’ operations. Currently, while the private sector often expresses opinions on popular short video apps like “Douyin”, the government and enterprises must invest considerable time and effort to retrieve information from such platforms, resulting in unnecessary resource waste. If information disclosure can be facilitated—such as through the establishment of more efficient platforms where government officials and firm representatives can directly engage with the public, or systems that gather, categorize, and organize public opinions collectively—the costs of public participation are anticipated to decline significantly. Consequently, both the public and firms would be more inclined to engage in policy implementation.

Thirdly, as we enter the new era of big data, it is suggested that policymakers leverage advancements in new technologies when interacting with the public. “E-government” or “e-governance” supported by the latest technology is believed to reduce bureaucratic red tape and enhance the operation of the long-standing traditional public policy system. For instance, text mining in big data analytics may help identify preferred items to report when evaluating policy feedback. It is essential to gradually transition to digital and smart governance to ensure accuracy and efficiency.

Fourthly, to cultivate meaningful and valuable feedback from the public, enhancing the general public’s literacy and awareness should be prioritized in the policy agenda. The public is unlikely to express concern or provide advice on corporate ESG performance if they are unaware of the term’s meaning. Policymakers can effectively utilize media for promotional campaigns or training purposes.

Fifthly, unlike citizens in some democratic cultures, Chinese citizens tend to be relatively introverted. Therefore, when voluntary public participation remains weak, incentives should be provided whenever necessary. For instance, one-time cash coupons could be distributed to citizens who offer opinions or participate in policy discussions.

Last but not least, when voluntary information disclosure fails, mandatory corporate information disclosure should be enforced, particularly for enterprises that have been involved in negative news. Regulations similar to the “Environmental Information Disclosure Measures” enacted in 2007 should be developed to compel enterprises to disclose critical information regarding corporate governance.

In fact, incorporating concerns of the general public into corporate ESG strategies is increasingly recognized as essential for building trust, securing loyalty, and ensuring long-term sustainability among internationally based SMEs. Ben & Jerry’s integrates social and environmental justice into its business model and often leans on public feedback to shape its policies. They have been assessing the environmental concerns of their customers through social media campaigns and feedback channels. In 2020, the company launched a campaign addressing climate change and committed to sourcing Fairtrade-certified ingredients, which was considered a direct response to consumer advocacy for ethical sourcing. This strategy facilitates customer engagement and strengthens the brand’s identity as socially responsible. As a result, Ben & Jerry’s has heightened its reputation and sustained a loyal customer base that prioritizes ethically sourced products.

Interface, a modular flooring company, has embraced sustainability as a core component of its business strategy. They launched the initiative named “Climate Take Back”, which aimed to not only reduce their environmental footprint but also to “reverse the impact” of climate change. Interface solicits feedback from its customers on sustainability efforts to form their ESG practices. Their “Net-Works” program, which recycles fishing nets to create carpet tiles, addresses consumer concerns about ocean plastic pollution. Incorporating consumer feedback has bolstered brand loyalty and positioned Interface as a leader in sustainability in the flooring industry. The company has witnessed significant growth in sales as a direct result of its environmental initiatives.

Incorporating public concerns into corporate ESG strategies is also increasingly relevant for SMEs in China, where consumer awareness of sustainability is rising. Beijing Tsinghua Tongfang Co., Ltd. (Beijing, China) is an SME that specializes in energy-efficient technology. Responding to public concerns about climate change and energy consumption, the company focused on developing smart energy management systems that optimize energy use in both residential and commercial spaces. They involved customers through workshops and feedback sessions to address specific environmental concerns. The company’s open line of communication with the consumers seeks their input while allowing them to participate in testing the technology. This achieved significant market adoption and led to increased sales.

While Haier is a large corporation, its initiative to support local SMEs in adopting sustainable practices is noteworthy. The Haier Eco-Design platform allows smaller manufacturers to access resources and gain support for developing energy-efficient and environmentally friendly products. The platform incorporates consumer feedback into the design process and offers environmental training for SMEs. By providing resources to SMEs, Haier has seen a surge in the number of energy-efficient products coming to market, helping both Haier and its partner SMEs acquire a more environmentally conscious customer base.

For Chinese SMEs, effectively incorporating public concerns into ESG strategies includes understanding consumer preferences, soliciting feedback, and adjusting practices accordingly. Companies may consider adopting surveys, social media, feedback sessions, etc., to facilitate consumer engagement. They may also establish platforms for idea exchange to lead discussions in the industry, influence other businesses to adopt similar practices, and foster an environment for mutual monitoring and learning. As public awareness of environmental issues continues to grow, SMEs that prioritize the public’s ESG considerations will likely find themselves well positioned for success.

4.2. The Driving Factors Behind the Impact of PEC on ESG Performance

We further explore the interaction effect of PEC and other driving factors on a firm’s ESG performance. Our goal is to assess whether changes in the selected variable influence the relationship between PEC and ESG. We analyze the interaction between PEC and ESG at different levels of the selected variable and compare these findings to the previously identified cyclical relationship pattern. This approach helps us determine the relevance of firm diversity in how PEC impacts corporate ESG initiatives and identify the most effective PEC value for diverse firms. Furthermore, we aim to explore the effect of these selected variables on ESG performance, independent of PEC’s influence. To achieve this, we study the relationship between the selected variable and ESG at varying levels of PEC. A total of ten variables are incorporated into the analysis, and the regression results are elaborated as follows.

4.2.1. Firm-Specific Indicators

Firm Size

First, we evaluate whether firm size matters when PEC impacts a corporation’s ESG performance. As shown in Figure 2, the cyclical correlation between PEC and ESG scores persists across various firm sizes, with this correlation being relatively stronger for smaller firms. Conversely, larger firms generally perform better in ESG, as the ESG score increases monotonically with firm size at any given level of the PEC index. This positive interplay between firm size and ESG score is particularly evident when the PEC index value is extremely high (public concern heightened).

It is believed that smaller firms, with limited resources and low market share, may be more vulnerable to idiosyncratic shocks and thus experience greater volatility in their business environment. While public attention serves as a valuable form of oversight, it also represents a particular type of external risk that the firm cannot easily manage. When there is significant public interest in firms’ sustainable practices, smaller companies may find it challenging to swiftly and adequately reallocate resources to meet the demands of an overly concerned public.

Firm Age

The second firm-specific indicator we investigate is firm age. As shown in Figure 3, the nonlinear cyclical correlation between the PEC index and ESG score is observable across different levels of firm age, further validating our previous conclusion. This correlation is more pronounced among younger firms, which, like smaller firms, are expected to experience greater uncertainties throughout the business cycle due to a lack of resources and networks. Therefore, they are likely to be more easily affected by public attention than older firms.

Conversely, ESG scores tend to rise with firm age when the PEC value is between 180 and 370 or exceeds 800. Importantly, Figure 1 shows that ESG scores decline as the PEC index increases within these same ranges. This implies that while many firms generally struggle, older firms are often better prepared with strategies to respond to early public scrutiny or manage significant external pressure.

SOEs or POEs

State-owned enterprises (SOEs) and privately owned enterprises (POEs) differ in how public concern affects their ESG performance, as illustrated in Figure 4. Generally, the ESG performance of SOEs fluctuates more significantly across various levels of PEC. However, SOEs exhibit more promising ESG performance only under moderate public attention (PEC index values ranging from 200 to 800), while at extremely low or high levels of PEC, POEs outperform SOEs.

As SOEs represent the image and reputation of the government, it is crucial for them to uphold high standards of conduct and transparency. Moreover, with the government as the controlling stakeholder, SOEs are accountable for achieving mandated policy targets. Therefore, when the public actively engages in corporate decision-making, SOEs are expected to outperform POEs as they receive government backing during these times. Conversely, at low levels of public concern, the government may not prioritize SOEs’ ESG development among other policy mandates. If public attention becomes excessively concentrated, on the other hand, the government may struggle to effectively respond to the demands of a more critical public, leading to worsened ESG performance.

Proportion of Independent Directors

The effect of PEC on a corporation’s ESG performance remains robust when the proportion of independent directors varies, and the impact is more pronounced when this proportion is relatively low. Additionally, holding the PEC index value constant, Figure 5 shows that the curve depicting the relationship between the proportion of independent directors and the firm’s ESG performance does not exhibit clear patterns when public curiosity in the firm is low (PEC index below 370). However, at medium to high levels of public concern, a larger proportion of independent directors significantly enhances corporate ESG performance.

This can be explained by the fact that independent directors typically maintain impartial perspectives during corporate decision-making and prioritize a holistic view of corporate performance, which encompasses the company’s diverse effects on society. Unlike some other directors, they do not concentrate exclusively on financial stability. As a result, independent directors are anticipated to contribute positively and significantly to the advancement of a company’s ESG initiatives. However, this necessitates that the public is sufficiently concerned about the issue for it to be prioritized in corporate strategic planning.

4.2.2. Firm Stability Indicators

Leverage Ratio

It is reasonable to predict a negative relationship between ESG scores and a firm’s leverage ratio at any given PEC index value, as illustrated in Figure 6. This is due to the volatility a firm may experience when its daily operations are largely financed by external funding.

Firms’ investment decisions and activities for sustainable practices may also be restricted if their debt levels remain abnormally high, as a greater portion of the firm’s cash flow must be directed toward servicing debt rather than implementing ESG initiatives.

Additionally, companies with high leverage may face negative perceptions from external stakeholders, which could hinder investment from those who prioritize sustainability and therefore dampen the companies’ ESG development.

Fixed Asset Ratio

In contrast to leverage ratios, Figure 7 shows that a higher fixed asset ratio improves corporate ESG performance when the PEC value is extremely low or high. However, when the PEC index remains at a moderate level (380 to 800), the fixed asset ratio and ESG score are negatively correlated.

We propose that the rationale behind this relationship is similar to our earlier findings regarding firm size and age. Firms generally prefer a higher proportion of fixed assets for their operations, as those with a greater share of fixed assets are better positioned to manage external pressures from the public.

Moreover, a robust fixed asset ratio indicates that the firm is making long-term investments, some of which may support sustainable operations, such as energy-efficient facilities. However, a larger proportion of fixed assets compared to current assets may signal less liquidity. Given that ESG practices typically emphasize long-term investments over short-term gains, yet they require effective monitoring and timely responses to policy changes, we believe there exists a trade-off between stability and liquidity that leads to the contradictory correlation between fixed asset ratios and ESG scores across various PEC index levels.

4.2.3. Firm Profitability Indicators

Return on Assets (ROAs)

Figure 8 indicates that, in general, ESG scores initially decline and then rise as the ROA value increases at a constant level of public concern. The turning point is observed at an ROA value close to zero. Companies that exhibit both the highest ROA and PEC index values tend to excel in their ESG performance.

On the other hand, Figure 9 depicts the joint influence of PEC and ROE on ESG outcomes. At any specific PEC index level, ESG scores decrease with negative ROE but increase as ROE turns positive. However, unlike the trends seen with ROA, we observe that firms with the lowest values for both ROE and the PEC index tend to achieve better ESG performance than their counterparts.

Return on Equity (ROE)

Both ROAs and ROE are indicators of financial soundness, yet they serve distinct purposes and can sometimes convey different narratives about a firm’s financial condition. It is reasonable to expect a company committed to ESG to invest in areas such as renewable energy infrastructure, sustainable manufacturing lines, and green technologies. These investments typically expand the asset base, and if the ROA value subsequently rises, it indicates efficient management of this newly formed capital for ESG practices. The greater the public concern, the more likely companies are to focus on acquiring ESG-related assets, which explains why a higher PEC index level amplifies the positive effect of the ROA value on a company’s ESG performance.

Regarding the combined effect of PEC and ROE on a firm’s ESG performance, we believe the justification aligns with our conclusion related to leverage ratios. While taking on more debt may decrease the proportion of equity on the balance sheet and lead to a higher ROE ratio, it may also be detrimental to the firm’s ESG initiatives, as previously discussed. This may explain why, as shown in Figure 9, a relatively lower ROE tends to be preferred when considering its impact on the firm’s ESG score.

Furthermore, when a firm suffers from negative ROE, it is crucial for the firm to quickly turn around this unfavorable financial situation to restore investor confidence. Demonstrating commitment to ESG development can enhance a company’s reputation and brand image, while positive public perception resulting from robust ESG practices can attract capital and talent, ultimately leading to revenue growth. Consequently, we find that firms with relatively lower ROE levels may be more proactive in managing their ESG practices.

Net Profit Margin (NPM)

Another indicator we are investigating is the net profit margin (NPM). Generally, an increase in NPM jeopardizes ESG performance at the same level of public concern, particularly after NPM transitions from negative to positive. This negative relationship is more pronounced when public concern remains relatively low, as illustrated in Figure 10. When NPM appears unpromising, the firm must prioritize sustainability and responsible governance as a means to improve profitability.

4.2.4. Firm Value Indicator (Tobin’s Q)

The last firm-specific indicator we examine is Tobin’s Q. As can be seen from Figure 11, ESG scores are observed to be positively correlated with this indicator, which is reasonable, as a higher Tobin’s Q (greater than 1) indicates a higher market valuation. Investors perceive greater potential for future growth in these firms, often resulting in increased demand for their shares. Consequently, these companies find it easier to secure capital, which ultimately supports ongoing ESG development.

4.3. Model Accuracy

To evaluate model accuracy, we computed the Mean Squared Error (MSE) of the traditional linear regression method (LRM) and our RAB approach, respectively. The error is found to be largely reduced with the machine learning technique. Specifically, as we utilize a Self-Organizing Map (SOM) to find the optimal model to run ANN that maximizes the improvement in the MSE, it is essential to balance the number of data observations in each model and the number of models to improve competitive learning (Cheong et al., 2022). Four models were chosen to run the ANN framework after several thousands of trial-and-error runs and

Table 2. Comparison of the MSEs between the LRM and ANN methods.

	LRM	ANN	Improvement (%)
Overall	0.907685	0.720027	20.67442
Model 1	0.912727	0.695811	23.76576
Model 2	0.985783	0.918827	6.792172
Model 3	0.855619	0.705819	17.50773
Model 4	0.902926	0.759431	15.89217

Source: authors’ calculation.

illustrates the improvement in the MSE based on the RAB approach relative to the traditional linear regression model.

Model 1 demonstrates the most impressive improvement in RAB’s MSE relative to LRM (i.e., 23.77%). The improvements in the MSE for models 2, 3, and 4 are 6.79%, 17.51%, and 15.89%, respectively. We further identify the ANN function for the chosen variables in each model and pool them into an overall LRM and ANN model to rerun the data. The result displayed in the “Overall” row shows a significant overall improvement in the MSEs of 20.67%. In conclusion, our SOM-ANN framework indicates a consistently significant improvement in explaining the empirical relationships among the variables concerned, as illustrated by the reductions in the MSEs across all four models.

5. Conclusions

This paper presents a pioneering investigation into the nonlinear and even cyclical interplay between public environmental concern (PEC) and corporate environmental, social, and governance (ESG) performance utilizing advanced machine learning techniques, specifically, the Self-Organizing Map–Artificial Neural Network (SOM-ANN) framework. Our findings reveal that PEC significantly influences ESG scores. With notable thresholds, corporate ESG initiatives exhibit varying levels of performance, depending on the level of public concern. An optimal PEC index of around 800 correlates with enhanced ESG performance, while excessive public scrutiny can lead to a plunge in corporations’ ESG scores, particularly for those struggling to adapt to heightened public expectations.

Furthermore, we find that seven firm-specific characteristics—such as size, age, ownership structure, governance practices, fixed asset ratio, leverage ratio, and Tobin’s Q—do not significantly alter the impact of PEC on ESG performance, as the cyclical relationship between PEC and ESG scores persists across various levels of the aforementioned firm performance indicators. Nevertheless, we find that smaller and younger firms are more sensitive to public concern while larger and older firms are better equipped to face stringently heightened public scrutiny. State-owned enterprises tend to perform better in ESG than privately owned enterprises, but only under moderate levels of scrutiny. The proportion of independent directors plays a crucial role in enhancing ESG performance, particularly in the face of rising public expectations. A higher fixed asset ratio improves ESG performance only when public concern is either extremely low or high, while a high leverage ratio (value of Tobin’s Q) in general negatively (positively) affects ESG scores.

On the other hand, the analysis also reveals that some financial metrics, including return on assets (ROAs), return on equity (ROE), and net profit margin (NPM), significantly influence the interplay between PEC and ESG performance. Instead of observing the cyclical correlation, increasing public attention generally enhances the ESG performance of financially sound firms, while for financially poor firms, a low to moderate level of public concern is suggested for the sake of continuous sustainable growth. Higher profitability metrics are found, in general, to be complementary to better ESG practices.

The findings of our work underscore the critical role of public engagement in shaping corporate ESG strategies. Policymakers should prioritize mechanisms that facilitate public participation in environmental governance, ensuring that firms are held accountable to societal expectations. Additionally, this study suggests that firms must develop adaptive strategies to manage public scrutiny effectively, particularly in the context of evolving firm characteristics, environmental policies, and stakeholder expectations. Our application of the SOM-ANN framework demonstrates a significant improvement in predictive accuracy compared to traditional linear regression models, as evidenced by a reduction in the Mean Squared Error. This highlights the effectiveness of machine learning techniques in capturing the intricate relationship between the variables concerned.

Despite these contributions, our study is not without limitations. The analysis is based on a specific dataset, which may limit the generalizability of the findings to other contexts or regions. Additionally, while we explored various firm-specific factors, there may be other contextual variables influencing the relationship between PEC and ESG performance that were not accounted for in this study.

Future research could expand on our findings by examining the relationship between PEC and ESG performance across different industries and geographical contexts. Longitudinal studies could provide insights into how these dynamics evolve over time, particularly in response to significant environmental events or policy changes. Furthermore, exploring the role of digital platforms and social media in shaping public concern and corporate responses could yield valuable insights into contemporary environmental governance. Finally, investigating the interplay between corporate financial performance and ESG initiatives could enhance our understanding of the long-term sustainability of corporate practices in the face of rising public expectations.

In conclusion, this study enriches the literature on sustainable finance by illustrating the complex dynamics between public environmental concern and corporate ESG performance using cutting-edge machine learning techniques, while also highlighting the necessity for adaptive strategies and proactive public engagement in fostering sustainable business practices.