Professional Documents
Culture Documents
Shuangyan Li1, Muhammad Waleed Younas*1, R.M. Ammar Zahid2, Umer Sahil Maqsood1
1
School of Economics and Finance, Xi’an Jiao tong University, Xi’an, Shaanxi, China
2
School of Accounting, Yunnan Technology and Business University, Yunnan, P. R China
ABSTRACT
In the midst of the ongoing digital revolution, businesses are increasingly embracing artificial
intelligence (AI) to optimize their operations. This study aims to explore the adoption role of AI
in environmental, social, and governance (ESG) performance. Further, it investigates this
relationship with respect to different ownerships i.e., State owned enterprises (SOEs) vs non-State-
owned enterprises (non-SOEs) and central-SOEs vs local-SOEs. By analyzing a sample of Chinese
A-share listed companies from 2010-2020, findings reveal that the adoption of AI significantly
improves firm ESG performance, underscoring the importance of technological advancements.
Furthermore, the impact of AIA on ESG performance is more pronounced in SOEs compared to
non-SOEs and in central-SOEs than local-SOEs. These findings suggest a strong involvement of
the Chinese government in driving the digital economy and its role in ESG initiatives. Finally,
study findings are robust and consistent across multiple statistical horizons, including 2SLS, PSM,
SysGMM, and instrumental variable analysis.
The Artificial intelligence adoption (AIA) is a remarkable innovation which reshaping our world
(Yu et al., 2023). It brought fundamental transformation in various domains of life and unlocked
limitless possibilities. For instance, AIA has revolutionized teaching and learning methods, making
education more accessible and personalized (Zhang & Aslan, 2021). Further, at corporate level it
enhanced management control, empowering organizations to make informed decisions and
optimize their operations (Berente et al., 2021). This widespread adoption of artificial intelligence
(AI) signifies its growing importance and potential to shape the future remarkably. The figure 1
depicts that AIA have been consistent and improving progressively over the years.
Figure 1 Provides the mean Value of AIA in Total Firms, SOEs, and Non-SOEs"
Existing studies examining the impact of AI technologies on firm performance which can be
categorized into two main strands. The first strand focuses on the tangible benefits of AIA, where
firms invest in utilizing AIA applications such as autonomous vehicles and predictive maintenance
to improve their output and operational efficiency (Brynjolfsson et al., 2018). The second strand
examines the intangible benefits of AIA, exploring how AI-based techniques contribute to a firm's
economic growth (Aghion et al., 2018), productivity (Czarnitzki et al., 2023), firms financial
performance (Bosse et al., 2023), and innovation (Bahoo et al., 2023). These studies provide
insights into how enterprises adapt the digital era, leading to significant changes in their
operational structure and financial performance. However, there are several environmental, social
and governance (ESG) issues that these companies need to address. These include weak
environmental awareness, insufficient attention to employee rights and conditions, and imperfect
product safety assurance systems (Wang et al., 2022).
Next, we analyze how AIA affects ESG performance based on the ownership structure of firms.
Previous studies have yet to explore the impact of state ownership on a firm's ESG prospects for
AIA. This study investigates the relationship between state ownership and ESG outcomes by
categorizing ownership into two main groups: state-owned enterprises (SOEs) and non-state-
owned enterprises (non-SOEs). The goal is to fill the research gap and understand how different
types of ownership may impact ESG performance. The distinction between SOEs and non-SOEs
has become increasingly significant in the current global economy. It can be seen a growing
number of SOEs may be noticed in the make-up of the Fortune Global 500 (FG500) firms. The
FG500 had 27 SOEs in 2000, but by 2017, there were 102, accounting for 20% of the FG500.
Notably, out of the initial 27 FG500 SOEs in 2000, 9 were from China. However, in 2017, 75 out
of 102, the majority were from China. Furthermore, China has a significant presence of SOEs on
a global scale, with approximately 150,000 SOEs, out of which 75 are listed in the FG500. China's
1978 economic reform accounted for this growth in SOEs and their market valuation. As a result,
China's SOEs have become vital players in both the domestic and global markets.
Consequently, the Chinese market is frequently the primary focus of academic study on SOEs
(Jiang & Kim, 2020; Lu & Zhu, 2020). As an extension of the government, state-owned enterprises
greatly affect the firm's objectives, CEO compensation, resource allocation, transparency, and
overall firm performance (Lin et al., 2020). Considering the managerial (the managerial
perspective focuses on the decision-making and strategic choices of company managers) and
political (refers to the influence of government policies, regulations, and political agendas)
perspectives, we contend that the state ownership of enterprises can play a significant role in
shaping the connection between artificial intelligence adoption and ESG performance. Firstly, the
operation of SOEs within specific regulatory frameworks and government policies provides a
structured environment for decision-making and concentrates on innovation, enabling them to
focus on important objectives, driving economic growth, ensuring political stability (Kowalski et
al., 2013), and driving SOEs towards enhancing profit maximization (Choe & Yin, 2000). This
shift towards profit maximization potentially leads to a greater emphasis on ESG considerations
and creates stronger incentives for SOEs to adopt AI technologies that drive sustainability.
On the other hand, non-SOEs, due to their limited political connections, higher information
asymmetry, and managers' compensation structures, may face challenges in generating high profits
This study obtained financial data from the China Stock Market and Accounting Research
(CSMAR) database. Information on ESG disclosure was obtained from the Bloomberg database,
and artificial intelligence data were obtained from the database (CSMAR). This centralized
database focuses solely on China's economic and financial matters. This database contains
comprehensive financial information about various companies listed in China's stock market. To
ensure the validity of our research, we carefully selected the companies to be included in our
analysis. We excluded companies in the financial and insurance sectors and those that received
special privileges or underwent transfers. Additionally, we excluded firms where essential
information was missing (Hao et al., 2023). After applying these criteria, the final sample included
4348 firm-year observations.
Our sample consists of both state-owned enterprises (SOEs) and non-state-owned enterprises (non-
SOEs) across different industries in China. We analyzed the financial data from 2010 to 2020. This
particular time period was chosen because, starting in 2010, Chinese A-listed firms began
voluntarily disclosing information related to environmental, social, and governance (ESG) factors.
By considering this period, we aimed to ensure that our findings would be reliable and unbiased
when assessing the impact of the COVID-19 pandemic (Broadstock et al., 2021).
Note: In Figure 2, the solid lines represent the direct impact of Artificial intelligence adoption (AIA) on
firm ESG performance. On the other hand, the dotted lines depict the moderating effect of types of State-
Owned Enterprises (SOEs) (local vs. central).
We gathered data on ESG from two main sources from Bloomberg ESG indices and the China
Stock Market & Accounting Research (CSMAR) database. Bloomberg relies on various publicly
accessible sources such as annual reports, sustainability reports, and company websites. In order
to have an accurate panel (Wang & Juo, 2021), we excluded firms with missing data. The
Bloomberg ESG indices utilize a comprehensive index system and encompass various original
variables related to emissions, climate change policies, and energy consumption. The environment
dimension of the indices includes factors such as air quality, ecological impact, energy usage,
waste management, and water resources. The social dimension covers aspects like community
relations, diversity, ethical compliance, employee well-being, and supply chain practices. The
governance dimension encompasses areas such as audit oversight, board composition,
compensation practices, diversity, independence, governance oversight, sustainability governance,
In this study, the main focus revolves around integrating artificial intelligence adoption (AIA)
within corporations. However, accurately measuring the degree of AIA implementation within
these firms has proven challenging for researchers and scholars. While a few studies have
successfully used textual analysis to assess AI adoption in organizations (Alekseeva et al., 2020;
Babina et al., 2020), this particular task has caught our attention. The reason is that listed firms
often disclose information about their AIA initiatives through textual disclosures within the
management disclosure and analysis (MD&A) sections of their 10-K filings in annual reports. The
MD&A section provide stakeholders valuable insights into a company's internal and external
operations. It is reasonable to expect that if a corporation is pursuing an AIA initiative, it will
disclose it in its MD&A. As a result, this study leverages machine learning techniques and Python
software to comprehensively evaluate the extent of AIA integration in firms by analyzing their 10-
K filings from annual financial reports using textual analysis.
The process involves obtaining financial reports from the CSMAR database for A-share listed
firms spanning 2010 to 2020. Firstly, we obtain financial reports from the CSMAR database for
these firms. We use a Python crawler function to extract the relevant information to access the
management disclosure and analysis (MD&A) section from the annual 10-K filings. Subsequently,
Numerous factors related to firms, industries, and boardrooms have been identified as potential
influences on ESG, as highlighted in prior studies (Ren et al., 2022; Romano et al., 2020). To
maintain consistency, we consider these factors as control variables that may impact ESG. These
variables include firm size (F_SZ), returns on assets (RO_A), Tobin's Q (TQ-ratio), cash flow
(CFLO), debt-to-assets ratio (DEBTA), number of board members (NoBM) and the tenure of
CEOs (CEOT). Thus, we incorporate these control variables in this study to address potential
agency costs. Specifically, we consider board size (BD_SZ), board independence (BO_IND), and
largest shareholder percentage (HH-Index) (Chow et al., 2018), as essential factors. Additionally,
we examine CEO characteristics as control variables with ESG investment decisions. This includes
variables such as retiring CEOs and CEO duality. CEO duality is represented by a dummy variable
called Duality (DUAL), indicating whether the CEOs and chairpersons are the same person. For
further details and explanations of these variables, see Table 1.
NoBM Number of board The total number of board meeting in one calendar year.
members
RO_A Return on assets Net income divided by total assets
BO_IND Board independence Board Independence is calculated as the number of
independent directors/total number of board members.
Board size Board size is calculated as the total number of directors on the
BD_SZ
board
The Herfindahl-Hirschman Index is a well-known indicator that
Herfindhal-Hirschan helps gauge a market's concentration within a specific industry.
HH-Index
Index It acts as a widely accepted tool to evaluate the level of
competition prevailing in the market.
This financial indicator is calculated by adding the value of
TQ-ratio Tobin's Q ratio stocks and the value of debt and then dividing the sum by the
total value of assets.
DEBTA Debt to assets ratio Total debt divided by total assets
CFLO Cash flow Log of cash flow
Duality A binary variable for CEO duality, taking 1 if the CEO also
DUAL
served as Chair of the Board, 0 otherwise
Retiring CEO A dummy variable takes the value 1 if CEO is retired,
CEORT
otherwise its zero.
CEO-TN CEO tenure Total Number of a CEO serving years
Moderator variable
State-owned A dummy variable equal to 1 if firm controlled by
SOEs
enterprises government, otherwise zero.
Local state-owned A dummy variable equal to 1 if firm controlled by local
Local_SOEs
enterprises government, otherwise zero.
Central state-owned A dummy variable equal to 1 if firm controlled by central
Central_SOEs
enterprises government, otherwise zero.
To examine how artificial intelligence (AIA) affects Environmental, Social, and Governance
(ESG) factors, we use the baseline model to test our hypothesis 1.
Where in equation (1): The subscript refers to each individual firm, while 𝑡 denotes the specific
year of observation. 𝐸𝑆𝐺!" represents the factors and their impact on each firm's performance, and
𝐴𝐼𝐴$% indicates the firm's utilization of artificial intelligence. The variable 𝑋!" represents a set of
control variables at the firm, industry, and board levels, including factors like firm size (F_SZ),
return on assets (RO_A), Tobin's Q (TQ-ratio), cash flow (CFLO), Herfindahl-Hirschman Index
(HH-Index), board independence (BO_IND), board size (BD_SZ), and CEO tenure (CEOT) (see
table 1 for more detailed information on these variables). Furthermore, 𝜔'()* is a vector
representing time-fixed effects, and 𝜔+,-./"*' is an industry-fixed effects dummy variable used to
address potential unobserved influences in the model. Lastly, 𝜀!" is the random error term. In order
to analyze the moderating impact of AIA on ESG from firm ownership (SOEs vs non-SOEs). We
make a simple adjustment to the model by incorporating an interaction term. The modified
regression equation is as follows:
𝐸𝑆𝐺!" = α + β# 𝐴𝐼𝐴$% + β& Ownership$% + β0 𝐴𝐼𝐴$% × Ownership$% + β0 𝑋!" + 𝜔'()*
+ 𝜔+,-./"*' + 𝜀!" 𝐸𝑞. (2)
Equations (2) and (3) illustrate the moderating influence of firm ownership and types of State-
Owned Enterprises (SOEs). The variable "ownership" encompasses both SOEs and non-SOEs,
while "TYP" represents the two types of SOEs: local and central. The comprehensive regression
model accounts for time- and industry-specific effects denoted by ωyear and ωIndustry, respectively.
The residual term is represented by εit.
5. Empirical Results
5.1 Descriptive statistics
ESG Env Soc Gov AIT F_SZ DEBTA CFLO BO_IND BD_SZ NoBM CEORT DUAL CEO-TN HH-Index TQ-ratio
ESG 1
Env 0.877*** 1
Soc 0.808*** 0.550*** 1
Gov 0.551*** 0.299*** 0.327*** 1
AIA 0.0619*** 0.0640*** 0.0306 0.0634*** 1
F_SZ -0.0747*** -0.0334* -0.00324 -0.217*** -0.0326 1
DEBTA -0.0898*** -0.0585*** -0.00900 -0.188*** -0.0413* 0.566*** 1
CFLO -0.0416* -0.0337* -0.00992 -0.0729*** -0.0184 0.135*** -0.0748*** 1
BO_IND 0.0717*** 0.0505** 0.0583*** 0.0605*** 0.0659*** -0.0525** -0.0279 -0.0282 1
BD_SZ -0.0648*** -0.0432* -0.0270 -0.122*** -0.100*** 0.248*** 0.134*** 0.113*** -0.521*** 1
NoBM 0.0603*** 0.0615*** 0.0436** 0.0405* 0.107*** 0.279*** 0.260*** -0.108*** 0.0511** -0.0291 1
CEORT -0.00615 -0.00235 -0.0246 -0.00173 -0.0212 -0.0131 0.00884 0.0313 -0.0185 -0.00732 -0.00867 1
DUAL 0.0696*** 0.0512** 0.0188 0.113*** 0.0264 -0.136*** -0.0798*** -0.0530** 0.104*** -0.165*** 0.0153 0.154*** 1
CEO-TN -0.0112 0.00911 0.00252 -0.0902*** 0.0734*** 0.0871*** 0.0245 0.0765*** 0.0668*** -0.00935 -0.0245 0.0662*** 0.169*** 1
HH-Index 0.0247 0.0157 -0.000824 0.0560*** 0.0734*** 0.0192 0.0572*** -0.111*** -0.00381 -0.0156 0.0728*** 0.0188 0.0411* -0.0992*** 1
TQ-ratio 0.0483** 0.0324 0.0359* 0.0591*** 0.132*** -0.276*** -0.206*** 0.0204 0.0318 -0.111*** -0.0275 0.00906 0.00910 0.0907*** -0.0895*** 1
* ** ***
Note: p < 0.05, p < 0.01, p < 0.00
The baseline regression indicates a positive association between AIA and the firm ESG
performance. However, concerns have been raised regarding the possibility of unobservable and
persistent firm characteristics driving the results, which can lead to biased estimates. We further
address these concerns through changes regression.
Changes regression involves analyzing the change in variables over time instead of their absolute
values (Zheng et al., 2023). This approach helps to control for unobservable and persistent firm
characteristics that remain constant over time. We conducted a regression analysis where we
regressed changes in ESG (ΔESG, ΔEnv, ΔSoc and ΔGov) on changes in artificial intelligence
measure (∆AIA) and other control variables. The results of this analysis are reported in table-5.
The results remain significant and positive with our baseline analysis.
In this section, we investigate whether the impact of AIA on ESG varies among different types of
State-Owned Enterprises (SOEs). SOEs serve as the primary extension of the Chinese government,
and they are categorized into central and local SOEs to ensure efficient operation. The central and
local governments are entrusted with the responsibility of overseeing these enterprises on behalf
of the state. Given the significant stake held by the central government in central SOEs, we focus
on analyzing the findings of local SOEs in columns 1-4 and central SOEs in columns 5-8 as shown
in Table 7. We observe that the coefficient of AIA for central SOEs is stronger and more statistically
significant at the 1% level in relation to ESG performance. On the other hand, the coefficient of
AIA for local SOEs is less pronounced, suggesting a weaker effect of AIA on ESG in comparison
to central SOEs. These results align with our hypothesis (H3). Additionally, the other control
variables in Table 7 exhibit expected signs and are statistically significant.
To ensure the reliability of our findings, we have employed three distinct methodologies, including
Propensity Score Matching (PSM), Two-System GMM (Generalized Method of Moments), and
Two-Stage Least Squares (2SLS) analysis, thereby validating the robustness of our results.
The innate differences in firm characteristics between SOEs and non-SOEs may affect the results.
We perform propensity score matching (PSM) to overcome this issue to match SOEs with non-
SOEs based on their observable characteristics. By using both PSM, we can draw definitive
inferences about the effect of AIA on corporate ESG without being affected by the innate
differences between SOEs and non-SOEs.
In order to assess the balance of covariates before and after the implementation of the treatment,
we first plot the distribution of the covariate of interest in Figure 4. This allows us to visually
inspect the differences in the covariate distribution before and after the treatment. Next, in Figure
5 (a) and (b), we use kernel matching to match the pre-treatment and post-treatment groups on the
covariates. Kernel matching involves using a kernel function to estimate the probability density of
the covariates in each group and then matching the groups based on their estimated probability
densities. This helps ensure that the treatment and control groups are comparable with respect to
the covariates and reduces the risk of bias in estimating treatment effects. Once we confirmed the
validity of our matched sample, we conducted a regression analysis on the matched data using
ESG as our variables. The results of this analysis are presented in Table 8, confirming our main
findings.
F_SZ
DEBTA
BD_SZ
CFLO
HH-index
CEORT
BO_IND
TQ-ratio
Unmatched
Matched
-50 0 50 100
Standardized % bias across covariates
2
4
1.5
3
kdensity _pscore
kdensity _pscore
1
2
.5
1
0
0
0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1
propensity scores BEFORE matching propensity scores AFTER matching
The study's main findings may be biased due to endogeneity concerns arising from reverse
causality. Firms that enhance ESG performance may be more likely to adopt AIA initiatives. To
address these endogeneity concerns, we reexamined the base regression model using 2SLS
estimation, a two-stage least squares approach. The artificial intelligence proxy (AIA_FUND),
which we used in this analysis as an instrument variable (IV), satisfies the exogeneity and
6.3 Conclusion
In this study, we examine the effects of artificial intelligence adoption (AIA) on Environmental,
Social, and Governance (ESG) practices, using principal agent theory, resource-based view theory,
and signaling theory as the foundation of our research. Through empirical analysis, we investigate
how AIA influences overall ESG performance and its three sub-dimensions: environmental
performance, social performance, and the quality of governance. By drawing on these theories, we
aim to gain a deeper understanding of the relationship between AIA and ESG practices in
organizations. Through an analysis of a sample of A-share listed companies from 2010-2020, our
findings demonstrate the significant impact of AIA implementation on improving firm ESG
performance. This underscores the importance of leveraging technological advancements to
enhance sustainability practices. Furthermore, our study reveals that the influence of AIA on ESG
performance is more pronounced in SOEs compared to non-state enterprises, indicating the
government's active involvement in promoting digitalization and sustainable development.
Specifically, the relationship is particularly strong in central SOEs, suggesting the central
government's role in driving the digital economy and fostering ESG practices. Importantly, our
study findings are robust and consistent across various statistical approaches, including 2SLS,
PSM, SysGMM, and instrumental variable analysis. These methodological robustness checks
provide confidence in the reliability and validity of our results. Overall, this study contributes to
the growing body of literature on the intersection of AIA and ESG performance. The findings
highlight the need for businesses to embrace AIA as a tool for driving sustainability and advocate
for supportive policies that encourage its adoption. Additionally, policymakers and government