ESG in the Financial Industry:

What Matters for Rating Analysts?

ed
Marco Mandas1,2 ; Oumaima Lahmar2; Luca Piras2; Riccardo De Lisa2

iew
Abstract

v
This paper investigates the main features highlighted by ESG (Environmental, Social and Governance) rating
analysts in the financial industry and the main differences across financial companies and sectors. We use a wide

re
database issued by Sustainalytics to perform a topic modeling and a sentiment analysis to identify the content of
analysts’ views on the companies’ ESG performance and to assess if the texts exhibit a positive or negative feeling.
The results suggest that for the financial sector, compared to the other ones, analysts’ views focus to a greater
extent on governance issues (Ethical conduct, ESG integration in the financing decisions, Product governance
transparency). The best financial entities in terms of ESG performance take greater care of data privacy and
product governance transparency. On the other hand, analysts call attention to the ethical and misselling conduct
er
held by the worst-scored financial institutions. We also find that the sentiment is coherent with the ESG scores.
pe
Keywords: ESG, ESG Rating Analyst, Banking, Topic Modelling, Textual Analysis, Environmental,
Governance, Social, Sentiment Analysis
JEL Classification: G20, G24, Q58
ot
tn
rin
ep

1 Corresponding author
2 Department of Economics and Business, University of Cagliari, Viale S. Ignazio 17, 09124 Cagliari, Italy
Pr

Marco Mandas: marco.mandas@unica.it

Oumaima Lahmar: oumaima.lahmar@unica.it
Luca Piras: pirasl@unica.it
Riccardo De Lisa: delisa@unica.it

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
ESG in the Financial Industry:

we
What Matters for Rating Analysts?

vie
Abstract

re
This paper investigates the main features highlighted by ESG (Environmental, Social and
Governance) rating analysts in the financial industry and the main differences across financial
companies and sectors. We use a wide database issued by Sustainalytics to perform a topic modeling
and a sentiment analysis to identify the content of analysts’ views on the companies’ ESG
performance and to assess if the texts exhibit a positive or negative feeling. The results suggest that
er
for the financial sector, compared to the other ones, analysts’ views focus to a greater extent on
governance issues (Ethical conduct, ESG integration in the financing decisions, Product governance
transparency). The best financial entities in terms of ESG performance take greater care of data
privacy and product governance transparency. On the other hand, analysts call attention to the ethical
pe
and misselling conduct held by the worst-scored financial institutions. We also find that the
sentiment is coherent with the ESG scores.

Keywords: ESG, ESG Rating Analyst, Banking, Topic Modelling, Textual Analysis,
Environmental, Governance, Social, Sentiment Analysis
ot

JEL Classification: G20, G24, Q58

tn
rin
ep

3
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
1. Introduction

we
This paper investigates how rating analysts form, express, and discuss Environmental, Social,
and Governance (ESG) ratings in the financial industry and what are the main determinants to
deserve a high ESG rating or getting a poor one.
ESG is an emerging stream of research that recently gained traction due to the importance of
these issues in the worldwide debate. Indeed, a growing number of institutions are disclosing their

vie
effort and accomplishments in ESG compliance, while ESG ratings and scores become a very
important factor in the overall performance evaluation.
Corporations and financial institutions are now held accountable for their environmental and
social footprint and to disclose and report on their activities. Hence, a large and diverse set of
standards, frameworks, and metrics related to what is referred to as ESG has been developed by

re
regulators and policymakers around the world. In terms of financial regulation, for example, the
Securities and Exchange Commission (SEC) in the US leaves it up to companies to determine what
is material information to be disclosed, while European authorities are implementing mandatory
obligations enforced by individual countries.
Institutions' ESG performance is becoming a growing focal point in financial markets where
er
investors decide and manage their investments also based on ESG criteria. The availability of ESG
ratings issued by rating agencies or, more generally, by specialized providers is a crucial factor in
mitigating information asymmetries and facilitating investors’ portfolio choices. Moreover, ESG
pe
performance is of growing interest for the goods and services market as consumers are paying
increasing attention to suppliers’ ESG performance in their buying decisions. Therefore, it is crucial
for companies and financial institutions to understand the determinants of a good ESG rating and
what actions prove to determine a poor or a good ESG rating.
The goal of this paper is to focus on what matters to ESG rating analysts in the financial industry.
In other terms, the main contributions of this work can be summarized in the answer to the following
ot

three research questions:

 What are the main features highlighted by ESG rating analysts?
 Do economic sectors exhibit significant differences from each other? And what are the
tn

characteristics of ESG ratings in the financial sector?

 What are the main issues highlighted by the analysts for the best and the worst financial entities
in terms of ESG performance?
rin

To answer these questions, this paper analyzes a wide database issued by Sustainalytics,
containing ESG ratings of more than 5,000 entities from 11 economic sectors. The Database contains
not only the ESG rating/scores of each entity but also the rating statements, in which the analysts
explain textually the main reasons for the assigned rating/score and other useful elements for
understanding the ESG performance (good or poor) of the entity.
ep

This paper, with a novel approach, focuses on companies' rating statements and performs a topic
analysis to investigate the content of analysts’ views on companies’ ESG performance. Topic
modeling is a machine learning technique that allows us to identify and cleanse the dominant topics

4
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
concerning ESG integration assessment, which makes it possible to sort companies based on their
ESG maturity (score/analysts judgment). We also perform sentiment analysis to assess if the texts

we
are more likely to stimulate the readers’ positive or negative feeling. Such an effort provides new
and diverse output, paving the way to a different approach to studying and interpreting them. Textual
data contains a very wide set of latent information seldom undetectable or maybe neglected, though
having great revealing potential, impossible to investigate otherwise.
ESG ratings/scores are synthetic assessments, mostly quantitative and numerical, whereas rating

vie
statements comprise also non-numerical and qualitative information such as ESG-related strategies
and activities.
Non-numerical information may embed latent contents such as concerns, sentiments, doubts,
qualified opinions, weaknesses, threat perceptions, consensus drivers, and many other important
features. The latter are used for the design and implementation of managerial actions, policy

re
strategies, and supervision effectiveness. All these would lead to proper conjunction between
“reality” and the “sense of reality”, typical of the language, as Bertrand Russel would have put it. In
other terms, the large existing literature on sentiment analysis and its effects on human decision
processes, though supported by rigorous methodologies, still leaves some room to further
speculation in order to shed light on the mechanisms through which sentiment affects human actions.
er
The adopted approach is based on textual analysis including two different methodologies
recently used in academic research: topic modeling and sentiment analysis. They are based on
computational operations applied to textual data that combine linguistics, semantics, computer
pe
science, and statistics to uncover latent information and sentiment in a text.
Our results suggest that analysts’ ESG views are well featured by 13 main topics that combine
differently across economic sectors and highlight significant sectoral diversity of ESG issues. The
focus on financial institutions highlights that this sector’s ESG ratings have specific features. The
results suggest that analysts’ reports focus to a greater extent on governance compared to the other
issues (Ethical conduct, ESG integration in the financing decisions, Product governance
ot

transparency).
Furthermore, the analysis allows distinguishing between the best and the worst financial entities,
in terms of ESG performance. Results show that the best-scored ones take greater care of data
tn

privacy and product governance transparency. On the other hand, analysts call attention to the ethical
and misselling conduct held by the worst financial institutions in terms of ESG performance. Thus,
it can be suggested that the more transparent the financial institution is, the more prone it is to reach
a higher ESG rating. As far as we are concerned, this is the first and a novel approach to investigate
ESG performance in the financial industry, using textual analysis integrated with quantitative
rin

information.
In the first section we present an extended literature review on ESG-related studies, followed by
some highlights on textual analysis contributions, particularly on finance-related texts, and
concludes with contributions on topic modelling. The following section describes the dataset and
ep

the methodologies applied to the study, followed by results and discussion.

5
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
2. Literature review

we
Over the past decades, the number of corporations disclosing detailed reports on their
“sustainable” activities and initiatives has been rapidly increasing (Roca and Searcy 2012). It comes
with no surprise that the focus is increasing on the added value of the textual information provided
by the disclosing entity which generates a research opportunity.
These disclosures, generally voluntary, describe managerial approaches, strategies, and

vie
decisions corporations opt for to implement and/or improve sustainability-related issues. In other
terms, these reports publicly share corporate initiatives and practices related to the implementation
of Environmental, Social, and Governance (ESG) criteria. The literature highlights a variety of labels
used to refer to these reports such as “ESG reports”, “sustainability reports”, “integrated reports”,
“corporate social responsibility reports”, “triple bottom line” etc. This variety can be explained by

re
the voluntary aspect of sustainability disclosure and the lack of standardization related to the type
of information disclosed. For simplicity, in this paper, we will refer to these solely documents as
ESG reports.
The increasing relevance and importance of ESG practices for governments, investors, and
stakeholders called the academic community for extensive, renewed, and deeper investigations. This
er
can be observed through the growing number of publications about ESG reports and the
methodological diversity applied to examine them. More than 2,000 studies investigated the
relationship between ESG activities and financial performance (Hubbard 2009; Friede, Busch, and
Bassen 2015; Revelli and Viviani 2015; Azmi et al. 2021). The debate about the nature of the ESG
pe
effect on financial performance is however not yet conclusive. A recent study developed what is
called the “ESG frontier” reconciling the adoption of ESG criteria on one side and the equilibrium
asset prices on the other side, attempting to empirically estimate the magnitude of ESG effect on
portfolio choice (Pedersen, Fitzgibbons, and Pomorski 2021).
Meanwhile, with the evolution of textual analysis, studies on corporate textual communications have
ot

been increasingly conducted by applying qualitative methodologies. In finance and accounting

research, the primary focus of the textual analysis was annual reports where researchers for example
evaluated the length of corporate disclosures using word count or file size (You and Zhang 2009;
Lang and Stice-Lawrence 2015). Some studies conducted readability analysis to evaluate the clarity
tn

and the understandability of corporate disclosures and their impact on investors’ behavior (Li 2008;
Loughan and McDonald 2014; Ibrahim and Hussainey 2019; Rjiba et al. 2021). Textual
communications proved to be powerful and influential on decisions, sentiments, perceptions, and
risk preferences which may explain the increase in the number of publications applying empirical
methods on document analysis in ESG research (Hahn and Kühnen 2013). As a matter of fact, these
rin

methods are applied to dig deeper into the qualitative data to obtain additional information not only
about the financial status of corporations but also to evaluate their commitment to ESG criteria.
Content analysis on corporate communications has been conducted to examine the nature and
the degree of disclosure, the influence of media on the information disclosed, and even to track
ep

changes in corporate releases over time. By establishing ESG checklists, word lists or dictionaries,
and categories, it is possible to score the disclosing entity and evaluate how mature it is in terms of
ESG compliance (Deegan, Rankin, and Tobin 2002; Jenkins and Yakovleva 2006; Giles and Murphy
2016; Baier, Berninger, and Kiesel 2020). This analysis is based on setting or listing ESG criteria
6
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
and determining their related words or dictionaries. Through checklists and word frequencies in a
document, it is possible to understand the importance of a particular criterion and the attention it

we
gets from the disclosing entity or stakeholders.
However, manually navigating ESG reports seems to be a daunting, time-consuming, and bias-
prone task which also results in a smaller number of examined data. More recently developed
techniques such as machine learning models applied to textual data facilitated the extraction of
information from a large corpus, which became a less complicated task, more reliable, and less time-

vie
consuming.
Topic modeling is one of the recently developed methodologies by which it is possible to obtain
the most relevant themes present in a textual document. The literature on topic modeling in finance
and more specifically on ESG reports are still scarce. As far as we are concerned, only few papers
used topic modeling approach on ESG reports.

re
Through topic modeling, Ning, Yim, and Khuntia (2021) tried to identify the users of ESG
reports based on the topics disclosed and their targeted readers. The authors argue that
information/topics embedded in ESG reports may be used to infer the strategic intent behind this
type of disclosure. Furthermore, they claim that the focus on ESG topics and themes to which a
er
corporation is committed may bring financial value. The paper also provides an insightful finding
about the strategic corporate intent of managing external relationships with stakeholders through
voluntary disclosure on ESG. In the same context another study was conducted to identify the
targeted readers of ESG reports based on the nature of the topics disclosed. The findings confirm
pe
the claim of Ning, Yim, and Khuntia, as firms adopt a stakeholder approach to management where
they are favored (Lindgren, Huq, and Carling 2021).
Another line of papers applied topic modeling for a descriptive purpose where the themes and
ESG criteria discussed by the Non-Financial Reporting Directive were the object of interest of the
authors. Fiandrino and Tonelli (2021) extracted four main themes that stood up in the discussion:
quality of NFI, standardization, materiality, and assurance. Székely and vom Brocke (2017) applied
ot

topic modeling on more than 9,500 corporate sustainability reports and extracted forty-two topics
linked to sustainability that are equally distributed among environmental, social, economic, and
general sustainability themes. Topics exposed under environmental issues are argued to be
tn

particularly linked to emissions and energy consumption. Parallelly, topics disclosed as a part of
social sustainability mainly refer to employees’ well-being and customer orientation.
Goloshchapova et al. (2019) on the other hand used topic modeling to compare the content of
voluntary disclosures between EU countries and the UK in terms of key concepts and criteria.
The literature on textual analysis applied to ESG reports and more specifically the one applying
rin

machine learning techniques is still nascent. In this paper, a new type of textual data, ESG analysts’
views, will be analyzed through topic modeling to understand the themes these texts discuss, and
extract the most dominant topics and their proportions across the corpus. To our knowledge, this is
the first study using topic modeling on analysts’ opinions about ESG performance to determine the
ep

major topics of interest and the criteria respected/violated by high/low scored corporations in terms
of ESG ratings.

7
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
3. Data and Methodology

we
3.1 Data
The paper uses the ESG Risk Rating Focus Database developed by Sustainalytics, a Morningstar
company. According to the company, ESG Risk Rating “captures an issuer’s exposure to the
material, industry-specific ESG risks and an issuer’s management of those risks. The ESG Risk
Rating provides investors with an overall company score based on an assessment of how much of a

vie
company’s exposure to ESG risk is unmanaged. The more of this risk that is unmanaged, then the
higher the ESG Risk Rating score” (Sustainalytics.com).
The database contains more than 100 variables of ESG performance for 12,480 entities, covering
a time span starting from Nov-2020 to Jan 2021. It provides several ESG scores, measuring selected
issues presenting material risks to entities’ performance. Overall scores are provided with qualitative

re
(textual) assessments, explaining company's ESG performance and its capacity to manage related
risks. These assessments represent the judgement for three material factors considered by the analyst
to be high priority for each company. Each analyst view forms a commentary on the company’s
performance with respect to the exposure and the management of the selected material ESG issue.
On average analyst views’ length is about 200 words each.
er
In this paper we select the entities having the three (qualitative) analyst views and the overall
(quantitative) ESG risk score. This leads to an ultimate selection of 5,012 entities, out of 12,480, for
Dec-20201 (Table 1). Companies belong to 11 sectors; the financial sector counts for 19% of total
pe
entities. Considering Continents breakdown, Europe has a weight of 31%, Asia/Pacific 35% and US
25%.
ot
tn
rin

Chart 1- Breakdown of the entities in the database divided by sector and origin (colours). The
y-axes reports the absolute number of the entities from the corresponding sector.
ep

1We select the textual data referred to a single period of time (Dec-2020) as the analysts’ qualitative assessment do not
change over the short time span that is covered in the database (from Nov- 2020 to Jan-2021)

8
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
we
Africa / Middle Latin America United States
Sector % Asia / Pacific % Europe % % % Total %
East and Caribbean and Canada
total Area total Area total Area total Area Total Area

Consumer Discretionary 7 1.18% 4.70% 223 37.7% 12.7% 181 30.6% 11.7% 12 2.0% 3.8% 169 28.5% 13.6% 592 12%

Consumer Staples 10 3.11% 6.71% 139 43.2% 7.9% 86 26.7% 5.5% 24 7.5% 7.7% 63 19.6% 5.1% 322 6%
Energy 7 2.62% 4.70% 68 25.5% 3.9% 76 28.5% 4.9% 26 9.7% 8.3% 90 33.7% 7.3% 267 5%
Financials 59 6.12% 39.60% 290 30.1% 16.5% 348 36.1% 22.4% 82 8.5% 26.2% 185 19.2% 14.9% 964 19%

vie
Healthcare 5 1.53% 3.36% 102 31.2% 5.8% 110 33.6% 7.1% 2 0.6% 0.6% 108 33.0% 8.7% 327 7%
Industrials 7 0.91% 4.70% 294 38.1% 16.7% 255 33.1% 16.4% 57 7.4% 18.2% 158 20.5% 12.7% 771 15%
Information Technology 6 1.24% 4.03% 199 41.2% 11.3% 95 19.7% 6.1% 5 1.0% 1.6% 178 36.9% 14.4% 483 10%
Materials 21 4.43% 14.09% 166 35.0% 9.4% 143 30.2% 9.2% 29 6.1% 9.3% 115 24.3% 9.3% 474 9%
Real Estate 14 3.43% 9.40% 153 37.5% 8.7% 102 25.0% 6.6% 29 7.1% 9.3% 110 27.0% 8.9% 408 8%
Telecommunication Services 9 7.09% 6.04% 44 34.6% 2.5% 47 37.0% 3.0% 16 12.6% 5.1% 11 8.7% 0.9% 127 3%
Utilities 4 1.44% 2.68% 79 28.5% 4.5% 110 39.7% 7.1% 31 11.2% 9.9% 53 19.1% 4.3% 277 6%

re
Total 149 2.97% 1757 35.1% 1553 31.0% 313 6.2% 1240 24.7% 5012 100%

Table 1 - Breakdown of entities by sector and origin. The table reports the absolute value of the
entities from a specific sector and region, the percentage with respect to the total number of entities
(column called “total”) and with respect to the total number of entities from the corresponding area
(column called “area”)
er
.

3.2. Methodology
pe
The analysis applied in this paper relies mostly on topic modeling. Topic modeling is a text
mining tool aiming at detecting and discovering latent topics discussed in a corpus which is, in this
paper, the analysts’ views database. In this context, we applied the LDA (Latent Dirichlet
Allocation) approach which is one of the many generative statistical models used in topic modeling.
LDA is defined as a dimensionality reduction technique attempting to discover the latent thematic
ot

structures in a corpus (Blei, Ng, and Jordan 2003).

The analysis follows a three-step process. The first step consists of applying topic modeling to
the entire set of analysts’ views related to entities in all sectors in the dataset. The aim is to extract
tn

an acceptable number of topics from the qualitative assessments made by the analysts. Secondly, we
focus on the financial sector by exploring in depth the distributions of each extracted topic and the
different features discussed by the analysts. Finally, a sentiment analysis is performed on analysts’
views to uncover the valence of the embedded sentiments related to the best and worst ESG
performing financial entities.
rin

Despite the automated aspect of topic modeling in analyzing a large corpus of textual data,
reliable analytical results require the corpus to be pre-processed and well-prepared for the analysis.
The preparation processes must be performed in such a way that only informative textual features,
relevant to topic modeling, are kept allowing the machine to work on them. The paper adopts the
ep

following processing pipeline by using the software R to prepare the corpus: i) cleaning the text
which involves different tasks (switch to lowercase, removing special characters, symbols, numbers,
and punctuation); ii) removing stop words such as “is”, “the”, “and” by using the list provided by
topic models package in R; iii) tokenizing the text; iv) lemmatizing the words.
9
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
Finally, we also remove some words that are extremely common in the analysts’ views: a list of
42 words2 that occur in more than 25% of all the documents. These words have limited

we
discriminatory power and do not help in classifying topics, preventing the model from estimating
them properly. Additionally, we also exclude all words that do not occur in at least 10 documents.
Quite the opposite, these words have too high discriminatory power as they may refer to a specific
company or a particular item and, consequently, are not useful for topic analysis.
Another methodological challenge is about the choice of the number of topics (K) to be extracted.

vie
Unlike other quantitative approaches, LDA is an automated machine learning tool, nevertheless, it
requires the number of topics (aka K) to be set in advance. Determining the number of topics a priori
is one of the issues of LDA, as users cannot know in advance the number of topics exhibited by the
corpus.
We opt for a trial-error approach, as there is no definitive answer to how many topics can be

re
extracted. After analyzing the results associated with a range of values of the number of topics K,
we decided to set the parameter to 13 topics, considering the quality of the output and the degree of
topic precision the model provided with K = 13. The choice of 13 topics is also supported by the
“perplexity metric”3 as shown in Chart 2. At K = 12 the perplexity score decreases abruptly and
then it starts to decrease at a slower rate.
er
pe
ot
tn

2The list of words that have been removed below: company, issue, average, management, exposure, risk, view, subindustry,
rin

medium, include, similar, practice, relate, low, product, program, address, moderately, measure, controversy, service,
implicate, preparedness, industry, business, system, increase, market, employee, strong, customer, impact, cost, minor,
expose, operation, lead, result, policy, regulatory, report, train.
3Perplexity is a statistical measure that is used to compare the generalization performance of models with different values of
K. The perplexity score is inversely proportional to the geometric mean per-word likelihood. The higher is the geometric
mean likelihood, the lower is the perplexity score and, eventually, the better is the generalization performance of the model.
ep

[Blei et al, 2003]

10
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
we
550

500
Perplexity Score

vie
450

400

re
10 15 20 25
Number of topics

Chart 2- Perplexity score

Once estimated the LDA model, the results have been analyzed in order to identify the label of
er
the topics. The labelling process has involved the following steps. First, we inspected the most
frequent words for each topic. Second, we use the LDAvis software, that is available in R, to further
investigate each topic by selecting the most relevant words according to the lambda measure4 set to
0.5. Third, we accurately read the analysts’ views presenting the highest percentage (this information
pe
is included in the Gamma matrix in the LDA results) to be about each topic. Finally, we explore the
terms’ concordances in order to identify in which context the top 10 most frequent words of each
topic5 are placed.
Having extracted and labeled the topics, the second phase of the analysis consists of an in-depth
analysis of two subsamples of entities, strictly belonging to the financial sector, comprising the
“Best” and “Worst” financial entities in terms of ESG score, respectively. To this end, the entities
ot

with the highest ESG scores (75% percentiles) and those with the lowest scores (25% percentiles)
are selected. On these two sub-samples, the most frequent topics were analyzed. The aim is to
understand which factors analysts focus on the most when assessing the ESG performance of the
tn

Best and the Worst financial entities.

The third and final task is aimed at identifying the analysts’ sentiment displayed in making their
judgments on Best and Worst financial entities. One wants to understand whether the previously
highlighted topics are presented by analysts with a positive or negative sentiment. We are, at this
rin

4 Lambda is the weight given to the probability of a term within a topic relative to the ratio of that term’s probability within
a topic to its marginal probability across the corpus. The higher is lambda, the higher is the weight given to the most frequent
terms within a topic. The lower is lambda, the higher is the weight given to the terms presenting a greater discrepancy between
ep

the frequency within a topic and the frequency within the corpus. We set lambda = 0.5 that is considered an optimal value to
compute the “relevance” of the terms.
5 See the Appendix to explore the entire list of concordances of each term.

11
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
point, able to find which are the most important factors, both positive and negative, that the analysts
highlight for Best and Worst financial entities.

we
In order to get the sentiment score, we use the polarity function in R software, which has a
built-in subjectivity lexicon (a predefined list of words associated with emotions or positive/negative
feelings). The function scans the text to identify words in the lexicon. It then creates a cluster around
an identified word. Within the cluster, valence shifters adjust the score. Valence shifters are words
that amplify or negate the emotional intent of the word. For example, "well known" is positive while

vie
"not well known" is negative. Here "not" is a negating term and reverses the emotional intent of
"well known." In contrast, "very well known" employs an amplifier, increasing the positive intent.
The polarity function then calculates a score using subjectivity terms, valence shifters, and the
total number of words. There are terms like “issue” and “risk” that are widely used in the analysts’
review that tend to lower the sentiment score. As a result, 80% of the reviews have a negative

re
sentiment score. Considering that such negative terms seem to be used in the same proportion for
each review, we could assume that the sentiment scores incorporate a constant negative bias.
For this reason, we normalize the scores to be between 0 and 1 by using the following formula:
zi = (xi – min(x)) / (max(x) – min(x)). Then we evaluate if the normalized values are above or below
er
the average.

4. Results
pe
The first results consist of 13 extracted topics from the corpus. The output of the topic modeling
is a set of keywords that have the maximum likelihood to represent each topic. Thus, based on the
most frequent word pattern of each topic and the concordances that these words have inside the
corpus, the topics are labeled. The labeling according to the literature is given based on the
researchers’ knowledge and intuition about the topic. In this paper, on the other hand, the labeling
ot

process is an educated “guess” based on semantics. After labeling the topics, we categorize them
into three major groups: topics related to environmental issues, social impact and governance aspects
that analysts expressed in their views.
tn

# ESG Label and most frequent words

1. E Sustainable energy
Carbon, Energy, Emission, GHG, Reduce, Regulation, Renewable, Fuel, Power, Gas
2. E Environmental and climate change risk
Emission, Waste, Community, Effluent, Environmental, Air, Project, Relation, Mine, Spill
rin

Sustainable supply chain

3. E Water, Supply, Resource, Right, Chain, Supplier, Material, Human, Production,
Manufacture
Human capital management
4. S Human Capital, Diversity, Manage, Initiative, Support, Workforce, Disclose, Growth,
Recruitment, Skill
ep

Human capital empowerment

5. S Human Capital, Labour, Development, Initiative, Turnover, Train, Talent, Diversity, Rate,
Develop, Experience

12
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
Data privacy and security
6. S Data, Security, Privacy, Information, Breach, Process, Cybersecurity, Sensitive, Financial,

we
Cyberattacks
7. S Consumer protection
Quality, Governance, Safety, Standard, Recall, Test, Ensure, Certification, Food, ISO
Health and safety
8. S Safety, Health, Occupational, Fatality, Incident, Contractor, Train, Accident, Performance,
Injury

vie
Business ethics and code of conduct
9. G Ethic, Corruption, Bribery, Ethical, Whistleblower, Anticompetitive, Bank, Fine, Tax,
Launder
10. G Product governance transparency
Disclose, Conduct, Major, Commitment, Access, Potential, Train, Lack, Initiative, Strategy
ESG integration in investment and financing decision

re
11. G Integration, Investment, Financials, Bank, Responsible, Opportunity, Asset, Project,
Diversify, Finance
12. G Real estate green finance
Build, Property, Real Estate, Revenue, Tenant, Green, Safety, Portfolio, Climate, Building
13. G Ethics conduct and misselling
Governance, Quality, Financial, Client, Bank, Offer, Lend, Control, Reputational, Poor
Table 2 - Topics labels
er
To further understand whether the financial sector shows different ESG features compared to the
other sectors, how the major three categories are distributed among them, we track the proportions
pe
of each topic in each sector as shown in chart 3.
Overall, the distribution of the topics is clearly sector driven. It is unsurprising that
environmental issues are strongly featured in sectors like Energy, Materials, Consumer Staples, and
Utilities. Particularly, the analysts highlight features related to sustainable supply chain for the two
consumer-oriented sectors (Consumer Discretionary and Consumer Staples). On the other hand, the
topics more related to carbon emissions, greenhouse gas production, and community integration
ot

(“Sustainable Energy” and “Environmental and Climate Change Risks”) are prevalent in the sectors
of Energy, Materials, and Utilities.
Focusing on the financial sector, topics related to environmental issues have a marginal weight,
tn

minimal when compared to other sectors, in analysts' views. Indeed, financial institutions play a
negligible role in directly affecting environmental aspects. However, they are crucial in determining
which criteria are relevant to have access to financing, which is significantly featured in the
governance-related topics about ESG integration. Furthermore, unlike environment-related topics,
governance issues attract analysts’ attention when it concerns financial institutions. In fact, the chart
rin

shows that the highest proportion of the extracted topics is governance related, featured by “Business
ethics”, “Product governance transparency”, and “Ethical conduct and misselling”. Precisely,
analysts discuss moral aspects such as anticorruption measures, commitment to transparency in
public disclosures and compliance with the code of best practice.
ep

Data privacy is also a matter of interest not only in the two technology sectors (Information
Technology and Telecommunication Services) but also in the financial one. Protecting sensitive data
is increasingly important as the analysts frequently warn financial institutions to pay attention to it.

13
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
As for the topics associated with social issues, the latter is more evenly distributed among the
different sectors. They play a major role in the Health Care sector. Indeed, compliance with high

we
quality and safety standards (“Consumer Protection”) is largely emphasized by analysts.
Human capital management and human capital-oriented policies are relevant for virtually all
sectors. They exhibit a significant appearance in Industrials, Consumer Discretionary, IT, and Real
Estate sectors.

vie
1.00

re
0.75

0.50
er topic
E − Sustainable Energy
E − Environmental and Climate Change Risks
E − Sustainable Supply Chain
S − Human Capital Management
S − Human Capital Empowerment
S − Data Privacy and Security
S − Consumer Protection
S − Health and Safety
G − Business Ethics and Code of Conduct
G − Product Governance Transparency
G − ESG Integration in Investment and Financing Decisions
pe
G − Real Estate Green Finance
G − Ethical Conduct and Misseling b y Financial Institutions

0.25
ot
tn

0.00

Financials Real Estate Telecommunication Services Healthcare Information Technology Industrials Consumer Discretionar y Consumer Staples Utilities Energy Materials

Chart 3 - Distribution of topics by sector

In order to respond to our research questions, we further investigate whether there are differences
rin

between the financial institutions having high ESG scores (Best) and low ones (Worst). Meanwhile,
the study also explores the nature of the sentiments embedded in the analysts’ views. Chart 4 shows
the size and the sentiment valence of each topic for both clusters of financial institutions.
The first insight is that the analyst’s views about the financial institutions with better ESG
performance emphasize the quality of their data privacy and security systems and their commitment
ep

to transparency. As illustrated in the chart, the “Product Governance Transparency” and “Data
Privacy and Security” topics are more relevant and are discussed by the analysts with a positive
sentiment for the Best-scored institutions. Protecting sensitive data and publicly disclosing

14
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
managerial and governance strategies are held in high regard by the analysts and turn out to be key
factors for getting a positive ESG score.

we
On the other hand, the second and most important finding is that financial institutions exhibiting
ethical issues and misselling conduct may be seriously penalized by ESG analysts. Accordingly, it
would be highly recommended to limit actions with bad reputational consequences in order to get a
higher ESG score. The chart below shows how analysts talk to a great extent about the topics
“Ethical conduct and misselling by financial institutions” and “Business ethics and code of conduct”

vie
with a very negative tone in reference to the worst financial entities.
These two results highlight how transparency is the way to increase the probability to obtain a
good ESG score. Hidden strategies, unethical behaviour, and lack of transparency are not favorably
welcomed by the analysts that tend to severely punish them. This evidence may be related to the
procedures followed for evaluating the ESG performance that are mainly based on public

re
disclosures, news, and non-governmental reports. The transparency in providing this information
establishes a sense of trust between the analysts and the disclosing entities. On the other hand, the
deficiency in communicating enough information about ESG integration increases the ambiguity
around the ESG performance of the entity and may significantly impact the rating and the analysts’
views.
er
Eventually, it is worth noting that the "S" factors related to human capital are relevant for both
clusters (Best and Worst), but more significant for the “Best”. Furthermore, the other topics related
to all environmental issues, some social topics (“Health and Safety” and “Consumer Protection”),
pe
and “Real Estate Green Finance” are less frequently mentioned by the analysts, as they might have
been considered to play a minor role in explaining ESG risk performance. The sentiment is still in
line with the ESG scores, that is positive for the Best and negative for the Worst.
ot
tn
rin
ep

15
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
Positive
G − Product Governance Transparency

we
Negative
Positive
G − ESG Integration in Investment and Financing Decisions
Very Positive
Moderate
G − Business Ethics and Code of Conduct
Moderate
Positive
S − Data Privacy and Security
Moderate
Very Positive

vie
S − Human Capital Management
Positive
Negative
G − Ethical Conduct and Misseling b y Financial Institutions
Very Negative
ESG_Rating
Very Positive
S − Human Capital Empowerment High Risk
Moderate
Low Risk
Positive
G − Real Estate Green Finance
Moderate

re
Positive
S − Consumer Protection
Negative
Positive
S − Health and Safety
Negative
Positive
E − Sustainable Energy
Moderate
Positive
E − Environmental and Climate Change Risks

E − Sustainable Supply Chain

Negative
er
Moderate
Moderate
0.00 0.05 0.10 0.15 0.20 0.25
Topic Size
pe
Chart 4 - Distribution of topics by “Best” and “Worst” Financial Institutions where the labels
next to each bar indicate the average sentiment score in a scale from “Very Negative” to “Very
Positive”

5. Conclusion
ot

The paper investigates the main ESG features highlighted by rating analysts in different
economic sectors. The main goal is to compare the financial sector to the other economic sectors in
terms of ESG features and to identify what drives the best and the worst ESG ratings.
tn

This study performs a topic modeling and a sentiment analysis to identify the content of analysts’
opinions on the companies’ ESG performance and to assess if the texts exhibit a positive or negative
feeling.
First, the analysis indicates that the breadth and complexity of the whole ESG ratings can be
rin

simplified to 13 main topics that combine differently in different economic sectors.

Overall, the distribution of the topics is sector driven. Environmental issues are featured in
industries like Energy, Materials, Consumer Staples, and Utilities. Social factors are, instead, more
homogenously distributed among sectors. Human capital management and human capital-oriented
ep

policies prove to be significant for all examined entities.

For the financial sector, compared to the other economic sectors, analysts’ ratings widely focus
on Governance factors (e.g., Ethical conduct, ESG integration in the financing and investment

16
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
decisions, Product governance transparency). In particular, topics such as ethical conduct,
anticorruption, and attention to transparency in disclosing are frequently discussed. Conversely, the

we
Environmental factors play a minor role in ESG rating.
The application of sentiment analysis to the selected topics allowed us to define ESG
performance profiles of the financial operators with the Best and Worst ESG scores.
The analyses show that the Best ESG financial players are mostly attentive to the transparency

vie
of their public disclosures about ESG integration, more precisely, entities that give particular
consideration to data privacy and product governance transparency. They also focus on Human
capital management and empowerment. On the other hand, the Worst ESG financial players have
strong issues with ethical and misselling conduct and information asymmetry.
Possible future avenues of research may involve replicating the analysis on different time periods

re
and samples to verify if and how selected topics evolve over time, eventually impacting assigned
ratings. Further, it could be verified, by examining banks’ public disclosures and sustainability
reports, whether the topics can give predictive indications of their possible ESG rating.

er
pe
ot
tn
rin
ep

17
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
6. Bibliography

we
Azmi, Wajahat, M. Kabir Hassan, Reza Houston, and Mohammad Sydul Karim. 2021. “ESG
Activities and Banking Performance: International Evidence from Emerging Economies.” Journal
of International Financial Markets, Institutions and Money 70:101277. doi:
10.1016/j.intfin.2020.101277.
Baier, Philipp, Marc Berninger, and Florian Kiesel. 2020. “Environmental, Social and

vie
Governance Reporting in Annual Reports: A Textual Analysis.” Financial Markets, Institutions &
Instruments 29(3):93–118. doi: 10.1111/fmii.12132.
Blei, David M., Andrew Y. Ng, and Michael I. Jordan. 2003. “Latent Dirichlet Allocation.”
Journal of Machine Learning Research 3(Jan):993–1022.
Cohen Priva, Uriel, and Joseph L. Austerweil. 2015. “Analyzing the History of Cognition Using

re
Topic Models.” Cognition 135:4–9. doi: 10.1016/j.cognition.2014.11.006.
Fiandrino, Simona, and Alberto Tonelli. 2021. “A Text-Mining Analysis on the Review of the
Non-Financial Reporting Directive: Bringing Value Creation for Stakeholders into Accounting.”
Sustainability 13(2):763. doi: 10.3390/su13020763.
er
Friede, Gunnar, Timo Busch, and Alexander Bassen. 2015. “ESG and Financial Performance:
Aggregated Evidence from More than 2000 Empirical Studies.” Journal of Sustainable Finance &
Investment 5(4):210–33. doi: 10.1080/20430795.2015.1118917.
pe
Giles, Olivia, and Daniel Murphy. 2016. “SLAPPed: The Relationship between SLAPP Suits
and Changed ESG Reporting by Firms.” Sustainability Accounting, Management and Policy Journal
7(1):44–79. doi: 10.1108/SAMPJ-12-2014-0084.
Goloshchapova, Irina, Ser-Huang Poon, Matthew Pritchard, and Phil Reed. 2019. “Corporate
Social Responsibility Reports: Topic Analysis and Big Data Approach.” The European Journal of
Finance 25(17):1637–54. doi: 10.1080/1351847X.2019.1572637.
ot

Guthrie, J., R. Petty, K. Yongvanich, and F. Ricceri. 2004. “Using Content Analysis as a
Research Method to Inquire into Intellectual Capital Reporting.” Journal of Intellectual Capital
5(2):282–93. doi: 10.1108/14691930410533704.
tn

Hubbard, Graham. 2009. “Measuring Organizational Performance: Beyond the Triple Bottom
Line.” Business Strategy and the Environment 18(3):177–91. doi: 10.1002/bse.564.
Ibrahim, Awad Elsayed Awad, and Khaled Hussainey. 2019. “Developing the Narrative Risk
Disclosure Measurement.” International Review of Financial Analysis 64:126–44. doi:
rin

10.1016/j.irfa.2019.05.006.
Lang, Mark, and Lorien Stice-Lawrence. 2015. “Textual Analysis and International Financial
Reporting: Large Sample Evidence.” Journal of Accounting and Economics 60(2–3):110–35. doi:
10.1016/j.jacceco.2015.09.002.
ep

Li, Feng. 2008. “Annual Report Readability, Current Earnings, and Earnings Persistence.”
Journal of Accounting and Economics 45(2–3):221–47. doi: 10.1016/j.jacceco.2008.02.003.

18
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
Lindgren, Charlie, Asif M. Huq, and Kenneth Carling. 2021. “Who Are the Intended Users of
CSR Reports? Insights from a Data-Driven Approach.” Sustainability 13(3):1070. doi:

we
10.3390/su13031070.
Loughan, Tim, and Bill McDonald. 2014. “Measuring Readability in Financial Disclosures.”
The Journal of Finance 69(4):1643–71.
Loughran, Tim, and Bill Mcdonald. 2011. “When Is a Liability Not a Liability? Textual Analysis,

vie
Dictionaries, and 10-Ks.” The Journal of Finance 66(1):35–65. doi: 10.1111/j.1540-
6261.2010.01625.x.
Ning, Xue, Dobin Yim, and Jiban Khuntia. 2021. “Online Sustainability Reporting and Firm
Performance: Lessons Learned from Text Mining.” Sustainability 13(3):1069. doi:
10.3390/su13031069.

re
Pedersen, Lasse Heje, Shaun Fitzgibbons, and Lukasz Pomorski. 2021. “Responsible Investing:
The ESG-Efficient Frontier.” Journal of Financial Economics 142(2):572–97. doi:
10.1016/j.jfineco.2020.11.001.
Revelli, Christophe, and Jean-Laurent Viviani. 2015. “Financial Performance of Socially
er
Responsible Investing (SRI): What Have We Learned? A Meta-Analysis.” Business Ethics: A
European Review 24(2):158–85. doi: 10.1111/beer.12076.
Rjiba, Hatem, Samir Saadi, Sabri Boubaker, and Xiaoya (Sara) Ding. 2021. “Annual Report
Readability and the Cost of Equity Capital.” Journal of Corporate Finance 67:101902. doi:
pe
10.1016/j.jcorpfin.2021.101902.
Roca, Laurence Clément, and Cory Searcy. 2012. “An Analysis of Indicators Disclosed in
Corporate Sustainability Reports.” Journal of Cleaner Production 20(1):103–18. doi:
10.1016/j.jclepro.2011.08.002.
Székely, Nadine, and Jan vom Brocke. 2017. “What Can We Learn from Corporate
ot

Sustainability Reporting? Deriving Propositions for Research and Practice from over 9,500
Corporate Sustainability Reports Published between 1999 and 2015 Using Topic Modelling
Technique” edited by R. Srinivasan. PLOS ONE 12(4):e0174807. doi:
tn

10.1371/journal.pone.0174807.
Tetlock, Paul C. 2007. “Giving Content to Investor Sentiment: The Role of Media in the Stock
Market.” The Journal of Finance 62(3):1139–68. doi: https://doi.org/10.1111/j.1540-
6261.2007.01232.x.
rin

You, Haifeng, and Xiao-jun Zhang. 2009. “Financial Reporting Complexity and Investor
Underreaction to 10-K Information.” Review of Accounting Studies 14(4):559–86. doi:
10.1007/s11142-008-9083-2.
ep

19
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
we
Appendix A: Topic top terms concordances

# TOPIC TERM CONCORDANCES

operation, exposure, company's, average, issue, management,
Sustainable Energy carbon
1 regulation, intensity, strict, risk
cost, renewable, source, company, efficient, increase, realize,
energy
short, transition, price
ghg, global, reign, wake, reduce, company, regulation, view,

vie
emission
progress, slow
emission, global, momentum, reign, wake, reduce, reduction,
ghg
view, progress, slow
emission, ghg, company, progress, slow, technology, energy,
reduce
initiative, electricity, process
change, climate, risk, relate, carbon, strict, view, company,
regulation
emission, associate
energy, source, transition, begin, power, increase, company,
renewable
generation, integrate, network
fossil, energy, company, source, emission, aircraft, combustion,
fuel
management, fuel, low

re
company, generation, renewable, coal, require, source,
power
continuous, supply, cable, exchange
emission, greenhouse, natural, increase, production,
gas
attributable, ghg, programme, world, fuel
reduction, deadline, ghg, emission, company, establish, reduce,
target
set, air, change
ghg, programme, emission, include, strong, target, risk, practice,
reduction
decommission, progressive
energy, change, result, company's, exposure, broad, risk,
price
regulation, financial, momentum
emission, ghg, momentum, political, reign, wake, 2, majority,
global

efficiency

climate
er
address, aircraft
cost, operation, opportunity, saving, energy, increase, emission,
company, crucial, tighten
change, regulation, risk, relate, term, company, increase,
associate, food, contextual
power, technology, electricity, coal, consume, network,
generation
renewable, significantly, attributable, heat
pe
energy, renewable, transition, begin, fuel, emission, ghg,
source
intensive, main, feedstock
carbon, trend, affect, business, energy, extent, model, company,
intensity
company's, configuration
energy, cost, emission, ghg, consumption, purchase, process,
electricity
benefit, decrease, future
Environmental and Climate waste, effluent, exposure, air, average, issue, company's,
emission
2 Change Risk management, excessive, community
emission, effluent, issue, company, view, produce, volume,
waste
mineral, similar, reduction
relation, project, exposure, average, issue, address, view,
community
scrutiny, company's, management
emission, waste, issue, management, view, average, company's,
ot

effluent
exposure, medium, address
management, system, certify, include, practice, impact, external,
environmental
industry, standard, due
emission, noise, excessive, community, trigger, vibration, water,
air
programme, waste, volume
community, delay, impact, protest, disruption, operation,
project
shutdown, significant, plan, address
tn

community, issue, average, company, company's, exposure,

relation
management, view, similar, moderately
produce, waste, amount, process, development, heavy,
mine
machinery, nox, site, sox
project, community, shutdown, mine, company's, exposure,
significant
impact, alter, landscape, generate

spill release, oil, contaminate, air, due, emission, fine, land, levy, nox

lawsuit, regulatory, severe, company's, exposure, regulation,

penalty
relevant, result, hundred, million
rin

unintended, spill, environmental, exceed, occur, regulator,

release
pollution, accidental, crude, due
hundred, dollar, cost, delay, company, develop, clean, include,
million
penalty, compensation
waste, produce, tailings, thicken, air, emission, report, company,
water
water, source
protest, project, permit, difficulty, dollar, million, community,
delay
industry, abandon, blockade
mineral, waste, mine, dust, heavy, machinery, nox, site, sox,
produce
amount
ep

leak, tailings, detect, mineral, waste, average, dewatering,

tailings
initiative, management, practice
monitor, community, identify, system, address, asset, concern,
stakeholder
key, life, management
environment, generate, impact, due, financial, risk, pipeline,
pollution
spill, subindustry, air

20
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
water, company, risk, efficiency, scarcity, clean, programme,
Sustainable Supply Chain water
3 management, comprehensive, include

we
chain, human, issue, exposure, management, view, low,
supply
company, disruption, risk
exposure, management, issue, average, company's, view,
resource
medium, water, risk, company
human, chain, supply, issue, exposure, average, company's,
right
management, abuse, association
supply, human, issue, company, view, low, moderately, risk,
chain
similar, implicate
audit, engagement, ongoing, train, regular, compliance, social,
supplier

vie
standard, set, comprehensive
raw, source, chain, produce, recycle, water, process, body,
material
lotion, sunscreen
supply, chain, issue, exposure, average, abuse, company's,
human
management, association, force
cost, increase, process, conflict, quantity, vast, growth, resource,
production
water, chemical
component, electronic, quantity, water, product, rely, region,
manufacture
defect, east, require
water, include, product, initiative, ban, develop, sale, efficiency,

re
reduce
reduction, cost
water, production, quantity, vast, manufacture, resource,
process
company, availability, clean, experience
hazardous, limit, substance, hazard, impact, risk, production,
chemical
harmful, involve, commodity
water, company, scarcity, scarce, prevalent, component,
region
electronic, manufacturer, company's, exposure
clean, electronic, component, manufacture, manufacturer,
component

recycle
er
quantity, water, activity, disruption, experience
water, reuse, drive, efficiency, reduce, company, invest, material,
minimize, system
hazardous, chemical, involve, regulation, worldwide, product,
substance
personal, act, company, consequence
component, manufacturer, issue, rely, experience, operational,
electronic
prevalent, region, company, manufacture
pe
company, preference, additionally, expose, shift, pressure,
consumer
regulator, subindustry, industry, cater
exposure, issue, management, average, company's, view,
source
medium, water, risk, company
management, issue, average, company's, company, exposure,
Human Capital Management human capital
4 view, inadequate, result, similar
employee, programme, demographics, embed, improve,
diversity
contribute, discrimination, limit, inefficiency, initiative
average, capital, human, company's, issue, exposure, view,
manage
medium, effectively, employee
ot

employee, programme, recruitment, retention, implement,

initiative
diversity, roll, train, tailor, company
drive, growth, innovation, strategic, workforce, depend, array,
support
complex, product, retain
growth, strategic, support, depend, retain, reits,
workforce
disproportionately, diversify, industry, technology
tn

employee, demographics, information, involve, practice,

disclose
average, diversity, alike, company, train
support, drive, innovation, strategic, workforce, foster, capital,
growth
deliver, human, inability
initiative, programme, retention, employee, diversity, embed,
recruitment
tailor, workforce, option, ownership
support, workforce, reits, company, design, estate, real,
depend
subindustry, industry, bank
rin

investor, view, alike, positively, reits, company's, exposure,

client
company, diversity, improve
deficit, highly, require, engineer, software, employee,
skill
compliance, failure, incentivize, develop
demographics, disclose, employee, information, practice,
involve
average, subindustry, construction, effectively, incentivize
support, drive, growth, strategic, foster, trust, loyalty, array,
innovation
complex, client
ep

compliance, programme, view, employee, company, diversity,

legal
improve, , ,
support, growth, innovation, strategic, foster, trust, complex,
drive
client, loyalty,
acquire, integrate, pose, risk, material, newly, respect,
talent
recruitment, board, human

strategic drive, growth, innovation, support, workforce, retain, strong, , ,

21
Pr

diversity, discrimination, limit, failure, inefficiency, operational,

improve
average, subindustry, initiative, access

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
Human Capital average, issue, exposure, management, company's, view,
human capital
5 Empowerment medium, strong, company

we
experience, company, relation, strain, require, represent,
labour
process, manufacture, skill, low
programme, human, capital, strong, company, include, talent,
development
initiative, engagement, opportunity
mentorship, wide, widespread, talent, diversity, development,
initiative
flexible, recruitment, workforce, labour
employee, impact, low, rate, form, line, bottom, motivate,
turnover
motivation, turnover
company, employee, labour, relation, include, activity, hr,

vie
train
experience, process, unmotivated
qualify, attract, recruit, employee, fail, innovation, stagnate,
retain
personnel, sound, strategy
include, initiative, development, average, practice, recruitment,
talent
attraction, plan, stock, talent
initiative, programme, workforce, labour, train, basic, flexible,
diversity
standard, monitor, employee
company, industry, seek, practitioner, product, skill, expose,
highly

re
personnel, issue, capital
average, capital, issue, subindustry, turnover, medium,
rate
employee, company, impact, recruit
human, capital, programme, strong, company, talent, employee,
develop
include, initiative, recruit
labour, company, relation, customer, diminish, insurance,
experience
manufacture, process, significantly, strain
retain, include, qualify, company, fail, employee, talent, option,
attract

staff
er
plan, practice
production, facility, staff, include, cost, company, compliance,
market, professional, sale
highly, require, industry, labour, low, design, manufacture,
skill
personnel, company, machinery
hour, relate, strike, freedom, company, experience, insurance,
wage
pe
restructure, fair, code
policy, association, freedom, level, minimum, salary, strong,
discrimination
labour, represent, workplace
industry, highly, skill, design, product, retain, key, airline,
personnel
attendant, mechanic
security, privacy, breach, exposure, issue, management, average,
Data Privacy and Security datum
6 company's, datum, company
datum, privacy, company, exposure, issue, view, management,
security
medium, risk, similar
datum, security, management, issue, company's, exposure, view,
ot

privacy
average, medium, risk
sensitive, identifiable, personally, datum, volume, security,
information
highly, customer, social, ensure
datum, increase, frequency, company, disruption, service, cost,
breach
total, loss,
tn

company, volume, store, identifiable, personally, receive,

process
security, define, ensure, information
exposure, due, regulate, sector, event, market, negative, react,
cybersecurity
share, typically
datum, information, identifiable, personally, customer, highly,
sensitive
market, react, access, industry
industry, datum, 2010, 3, experience, sector, record, credit,
financial
drive, history
rin

datum, exposure, frequency, breach, prime, security, target,

cyberattacks
collect, risk, industry
privacy, risk, security, infrastructure, technology, update, view,
assessment
regular, conduct, base
identifiable, information, sensitive, process, volume, datum,
personal
intellectual, property, decision, regard
datum, remove, system, damage, relate, repair, emerge, exist,
threat
ep

plan, address
datum, customer, company, behaviour, process, telecom,
collect
transmit, cyberattacks, store, complement
sector cybersecurity, due, regulate, 2010, 3, experience, financial, , ,
assessment, evolve, infrastructure, risk, update, company,
technology
proactive, hardware, issue, product
personally, information, datum, process, sensitive, volume,
identifiable
property, intellectual, addition, collect 22
Pr

cybersecurity, exposure, regulate, sector, breach, customer,

due
datum, dollar, million, ten
cybersecurity, due, sector, confidential, information,
regulate
manipulation, approach, base, datum, insurance
management, oversight, environment, system, datum, issue,
level
security, strong, assign, managerial

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
safety, management, product, machine, regulation, system,
Consumer Protection quality
7 standard, include, practice, determine

we
product, issue, average, company, company's, exposure,
governance
management, view, similar, medium
quality, product, machine, include, management, system, affect,
safety
directly, certification, implement
management, quality, expensive, extensive, lead, recall,
standard
regulation, international, iso, practice
product, lead, customer, complaint, expensive, increase,
recall
regulatory, scrutiny, standard, 9001
quality, 17025, adverse, asw, certify, control, division, iso, pass,

vie
test
reaction
company, quality, advance, machinery, stage, alignment,
ensure
regulatory, standard, manufacture, product
management, system, safety, 9001, iso, facility, accord,
certification
international, programme, ethical
administration, authority, drug, fda, safety, concern, lead,
food
producer, quality, issue
9001, certification, proactive, product, recall, system, certify,
iso
9000, additionally, company

re
adhere, deceptive, extensive, false, market, regulation, meet,
failure
regulatory, requirement, addition
clinical, datum, quality, machine, machinery, 3, complaint,
trial
customer, handle, transparency
ethical, incentive, market, personnel, regular, structure, train,
sale
company's, exposure, product
customer, company's, escalate, frequently, impact, affect,
lawsuit
compliance, cost, scrutiny, action
er
implement include, management, practice, quality, safety, system, , , ,
datum, trial, complaint, customer, handle, transparency, regard,
clinical
transparent, company,
effect, include, industry, irregularity, unanticipated,
manufacture
pharmaceutical, biotech, company, gmp, practice
threaten, customer, industrial, injury, result, safety, machine,
pe
defect
purchase, machinery, product
assessment, risk, periodic, product, weak, programme, evidence,
conduct
food, safety,
health, occupational, management, issue, company, exposure,
Health and Safety safety
8 view, include, practice, record
safety, occupational, management, average, issue, company's,
health
exposure, view, medium, train
health, safety, exposure, average, issue, company's, management,
occupational
disease, injury, illness
employee, limit, result, workplace, fatality, injury, morale,
ot

fatality
productivity, prone, commonplace
safety, investigation, temporarily, occupational, diesel, fuel, rate,
incident
lose, time, company
train, safety, employee, oversight, tend, fatality, lead, negative,
contractor
view, monitor
tn

safety, health, contractor, extend, oversight, employee, regular,

train
include, management, system
avoid, essential, integrity, maintain, productivity, workforce,
accident
damage, main, ship, financial
safety, poor, correlate, environment, financial, industrial,
performance
labour, record, strike, trigger
disease, financial, impact, occupational, illness, fatality, result,
injury
typically, industry, company
rin

safety, issue, average, subindustry, company, medium, key,

rate
environment, hazardous, employee
inefficiency, result, operation, tend, fatality, avoid, essential,
operational
integrity, response, preparedness
compensation, family, lawsuit, expose, ability, attract, exposure,
worker
company's, adequate, equipment
oil, nature, volatile, issue, key, petrochemical, company, natural,
gas
safety, contractor
ep

include, management, preparedness, risk, response, train, check,

emergency
medical, safety, system
safety, include, 18001, ohsas, health, maintain, alignment,
standard
management, certify, compliance
employee, train, safety, check, inspection, upgrade, assessment,
regular
equipment, risk, protective
safety, company, key, environment, hazardous, offshore, rig,
operate
employee, procedure, assessment 23
Pr

result, fatality, typically, lawsuit, low, morale, operation,

fine
shutdown, penalty, company
remote, environment, company, operate, chemical, contact,
hazardous
diesel, condition, include, perform

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
Business Ethich and Code of business, issue, infraction, company's, exposure, company,
ethic
9 Conduct average, management, view, include

we
bribery, risk, company, management, expose, company's,
corruption
exposure, issue, view, average
corruption, risk, company's, exposure, expose, average, issue,
bribery
industry, management, medium
include, infraction, broad, market, country, operate, insider,
ethical
practice, trade, commitment
report, confidentially, treat, programme, average, subindustry,
whistleblower
company's, unclear, anonymous, permit

vie
service, trustworthy, diversify, play, vital, fine, receive, record,
bank
claw, discipline
ethic, infraction, receive, record, bank, investment, lawsuit,
fine
expose, hit, million
evasion, compliance, launder, money, regulation, diligence, due,
tax
practice, account, competitive
money, sanction, violation, regulatory, regulation, ethic,
launder
international, link, programme, compliance
country, fund, movement, multiple, fraud, practice, relate,

re
account
taxation, tax, anti
ethical, include, ethic, fine, receive, record, broad,
infraction
manipulation, market, bonus
corruption, award, contract, expose, anti, lawsuit, practice,
investigation
relate, abuse, company
service, product, offer, complex, financial, bank, diversify, relate,
range
business, ethic
er
launder, sanction, violation, regulation, ethic, international,
money
link, programme, compliance, promote
operate, ethical, include, infraction, multiple, account, casino,
country
fund, vulnerable, develop
property, anti, competitive, relate, company, practice, issue,
intellectual
account, business, emerge
pe
complex, offer, product, range, service, effect, knock,
financial
performance, company, guilty
issue, insider, trade, expose, highly, surround, launder, money,
conflict
business, ethic
Product Governance company, price, list, publicly, policy, change, average,
disclose
10 Transparency programme, employee, product
risk, customer, framework, incorporate, insurer, train, code,
conduct
company, impact, social
controversy, implicate, issue, relate, access, barrier, company,
ot

major
disease, drug, federal
customer, profitability, compensate, insure, tension, honour,
commitment
honor, include, insurer, pay
service, basic, issue, average, exposure, management, company's,
access
ensure, medicine, company
tn

risk, company, price, scrutinize, 19, covid, exposure, legal,

potential
novartis, regulatory
santander, digital, platform, support, transformation, customer,
global
product, access, addition, bank
market, increase, budget, care, develop, health, public, system,
train
employee, conduct
company, pressure, increase, drug, product, adjust, provide,
price
biotech, coverage, regulatory
rin

disclosure, satisfaction, clarity, inform, measure, regulatory, risk,

lack
zurich, company, product
bn, usd, 1, 100, 300, amount, billion, programme, relate,
usd
settlement
improve, company, access, average, medicine, icbc, quality,
initiative
service, disclose, include
product, pursue, aggressive, financial, gain, compliance, explore,
ep

strategy
governance, regulatory, consumer
company's, exposure, fine, result, settlement, company, price,
significant
average, risk, issue
company, customer, price, product, expand, performance,
base
engage, pharmaceutical, care, health
result, hundred, penalty, company's, life, customer, eur, million,
million
company, fine 24
Pr

customer, ensure, liberalization, protection, rate, share, market,

china
access, china, drug
care, health, capacity, product, price, local, performance,
provide
agreement, build, public
complaint, report, market, offering, increase, 6, average, h1,
annual
resolution, semi
compliance, ethic, price, probe, bank, complaint, customer,
committee
dedicate, drug, establish

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928

ESG integration in
investment and financing
d
esg, financials, issue, average, company's, exposure,
integration
management, view, medium, address

we
11 decision
product, decision, esg, integrate, align, deliver, demand, depend,
investment
innovative, moral
integration, esg, issue, company, exposure, management, view,
financials
medium, similar, address
diversify, product, array, comprehensive, criticism, lend, level,
bank
offer, operation, subject
finance, opportunity, asset, bank, relate, risk, expectation,
responsible

vie
increasingly, public, shift
responsible, business, capture, array, relate, risk, multifaceted,
opportunity
asset, manager, finance
manager, management, opportunity, asset, increase,
asset
increasingly, responsible, company's, exposure, source
energy, lend, controversial, industry, efficiency, renewable, coal,
project
mine, likewise, company's
bank, product, array, comprehensive, criticism, lend, offer,
diversify
operation, subject, multifaceted

re
responsible, bank, opportunity, relate, expectation, increasingly,
finance
public, suggest, ,
energy, sustainability, esg, company, bond, green, renewable,
invest
cater, consumer, mind
product, bank, comprehensive, diversify, deliver, depend,
offer
innovative, investment, subindustry, client
environmental, challenge, tackle, business, equity, fund, asset,
social

environmental
er
esg, governance, integration
social, challenge, tackle, business, asset, manager, financials,
governance, integration, involve
asset, manager, finance, responsible, suggest, investor,
increasingly
millennial, term, client, offer
project, bank, controversial, encounter, footprint, geographical,
pe
lend
industry, operation, risk, diversify
project, encounter, industry, lend, risk, dam, hydroelectric,
controversial
manufacturer, mine, palm
carbon, estate, investment, low, portfolio, real, financial, align,
sustainable
business, development
integration, esg, issue, company, exposure, management, view,
financial
medium, similar, address
esg, decision, investment, customer, demand, evolve, bank,
integrate
consideration, reduce, efficiency
ot

green, certification, development, efficient, energy, system,

Real estate green finance build
12 premium, space, provide, safety

property risk, change, equip, ill, invest, affect, sale, develop, adversely, aid

company's, exposure, impact, reit, sale, adversely, affect, capital,

revenue
tn

company, devaluation
erode, finance, restriction, result, tight, esg, income, integrate,
real estate
manager
cost, increase, productivity, satisfaction, utility, worker, safety,
tenant
build, poor, ability
build, certification, development, efficient, energy, system,
green
building, premium, space, provide
tenant, quality, company, build, poor, compliance, legislative,
rin

safety
requirement, building, hurt
affect, asset, change, climate, estate, negative, performance, real,
portfolio
strand, dividend
management, factor, investment, process, risk, income, manager,
term
owner, reits, climate
change, estate, risk, additionally, regulation, invest, portfolio,
climate
result, tight, unitholders
ep

green, premium, space, provide, damage, likewise, vulnerability,

building
unitholders, company, quality
build, certification, comprehensive, green, robust, aid,
development
assessment, company, decision, property
esg, financials, issue, average, company's, exposure,
integration
management, view, medium, address

financials
esg, integration, issue, company, exposure, management, view, 25
Pr

medium, moderately, similar

assessment, esg, integrate, risk, estate, real, company, aid, build,
standard
construction
factor esg, include, platform, risk, term, decision, key, tenant, client,
change climate, property, risk, invest, portfolio, develop, estate, , ,
asset, negative, performance, portfolio, strand, lower,
affect
potentially, property, adversely, capital
annually, audit, compliance, legislative, property, quality,
ensure
include, initiative, practice, product

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928
 d
Ethics Conduct and product, management, issue, company, average, company's,
governance
13 Misseling exposure, view, refer, similar

we
management, control, internalize, product, service, financial,
quality
thousand, practice, system, branch
product, management, offer, quality, service, company,
financial
compare, impact, low, subindustries
interaction, individual, thousand, branch, view, company,
client
control, cross, transaction, favour
management, incentivizes, institute, oversight, practice,
bank
regulatory, governance, product, quality, refer

vie
offer financial, product, service, include, misconduct, type, , , ,
false, market, discriminatory, practice, predatory, misconduct,
lend
include, type, investor, mislead
internalize, quality, thousand, management, system, branch,
control
client, advisor, individual, interaction
demonstrate, operational, risk, bank, company's, exposure, loss,
reputational
mismanagement, company, controversy
disclosure, foreclosure, illegal, investor, lend, mislead, practice,
poor
consequence, crisis, demonstrate

re
poor, foreclosure, illegal, investor, mislead, practice, excessive,
disclosure
include, product, investment
average, institute, regulatory, subindustry, incentivizes, bank,
oversight
asset, manager, company, financial
include, misconduct, market, discriminatory, lend, offer, service,
type
false, illegal, model
include, type, market, lend, discriminatory, practice, offer, poor,
misconduct

fine
er
false, illegal
mismanagement, possibility, risk, expensive, penalty, company's,
,,,
lend, include, market, discriminatory, practice, predatory,
false
disclosure, misconduct, poor, type
governance, product, management, bank, asset, managers,
pe
refer
company, regional, requirement, service
client, interaction, thousand, branch, company, control, quality,
individual
,,
client, individual, branch, view, company, thousand, average,
interaction
control, ,
ot
tn
rin
ep

26
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4359928