Over the past decade, Directors and Officers (D&O) underwriters have observed the evolution of loss
causes leading to securities class actions (SCA). While in early 2010 almost all SCA settlements were
caused by activities related to core corporate governance, recent SCA settlements are indicative of a
variety of loss causes with a significant presence of environmental, social, and governance (ESG) risks as
underlying factors. Our review of the SCA settlements recorded between 2020 and 2022 suggests that
64% of settlements correspond to environmental and social risks, while the remaining 36% are related
to governance risks.

ESG-related regulatory requirements are on the rise in the U.S. and globally. The United States Securities
and Exchange Commission (SEC) has proposed rules to enhance and standardize climate-related
disclosures for investors requiring all public companies to disclose any potential climate-related risks
and their greenhouse gas emissions.2 It is anticipated that SEC rules will go into effect in 2024. Globally,
the European Union has been leading efforts in more stringent regulatory requirements to mitigate ESG
risks. The new German Supply Chain Due Diligence Act,3 which went into effect on January 1st, 2023,
requires that any German-based company with over 3000 employees report any Human-rights and
Environmental related risks in their supply chains.

With the increasing importance of ESG risks, underwriters are facing three key challenges managing the
impact of such risks on their book of business:

  1. Scalability: The process of extracting relevant ESG signals from the large body of reports
    and articles, also referred to as unstructured data sources, is costly and labor-intensive.
  2. Repeatability: With the continuous evolution of ESG risks and the corresponding
    regulatory requirements, underwriters need access to a repeatable insight extraction
    process that can be updated according to the latest definitions and regulatory
  3. Reliability: ESG insights have not been integrated into predictive analytics used by
    underwriters due to the above two limitations as well as the need to adequately
    calibrate such analytics and demonstrate that the changes are transparent, explainable,
    and consistent across policies and portfolios.

Consequently, in conversations with underwriters during recent Professional Liability Underwriting
Society (PLUS)4 conferences, we see that underwriters generally investigate and analyze ESG risks on a
case-by-case basis.5

In this article, we explain an analysis that offers practical ways to resolve the above pain points enabling
underwriters to consistently extract verifiable and scalable business values from relevant ESG insights
and in line with their long-term business growth strategy.

Our analysis of ESG risks for publicly traded companies suggests that commercially available ESG scores
often lack conformity and transparency.6 In further looking into this issue, we believe that reliable ESG
insights primarily reside in reports, news articles, and credible research studies, collectively referred to
as unstructured data sources, which are often not fully integrated into ESG scores. Processing
unstructured data sources manually is time-consuming and expensive. With recent advancements in
Artificial Intelligence (AI) capabilities in Natural Language Processing (NLP), unstructured data sources
can quickly be processed and relevant ESG insights extracted at scale and low cost. The key to extracting
high-quality ESG insights is to:

A. Ingest a diverse set of unstructured data sources.
B. Combine general and financial domain-centric Large Language Models (LLM).
C. Develop comprehensive taxonomies.
D. Extensively calibrate the NLP capability.

The Underwriting community has highlighted the need for more concise definitions of ESG risks. The
definition of these risks has been evolving over the past few years and we anticipate further updates
and refinements in the coming years. Consequently, underwriters need a solution that can be updated
in line with changes in the scope of ESG risks. In our analysis we adopted the following definition for ESG
risks: Specific actions taken by a company that can clearly lead to a reduction or increase of these risks.
We used a company’s balance sheet to define specificity and clarity in our definition. The table below
shows a few examples of how our analysis considers different sentences extracted from our
unstructured database for the companies we analyzed.

The solution architecture is designed in a way that the model can be updated without the need to
update the code and the outputs can be regenerated based on the updates made to the definition and

Table. Examples of Sub-Text identified by the model as relevant to ESG Risks as well as model forecast in terms of contribution to ESG Risks (Positive, Negative, Neutral)

Identified sub-textModel Output
The clinical trial of a cancer drug rejected by the Food and Drug Administration last week was marked by thousands of violations of study rules, damning investigator misconduct, and worrying signs of toxicity the company did not publicly disclose.Relevant, Negative
A Potentially Responsible Party (PRP) has joint and several liabilities under existing U.S. environmental laws. Where we have been designated a PRP by the Environmental Protection Agency or a state environmental agency, we are potentially liable to the government or third parties for the full cost of remediating contamination at our facilities or former facilities or at third party sites.Relevant, Neutral
We have evaluated the benefits of relying on other exemptions and reduced reporting requirements under the JOBS Act.Relevant, Positive
Our sales to the U.S. government are subject to extensive procurement regulations, and changes to those regulations could increase our costs. New procurement regulations, or changes to existing requirements, could increase our compliance costs or otherwise have a material impact on the operating marginsRelevant, Negative
We have faced and will continue to face claims relating to content that is published or made available through our products and services or third-party products or services. [From a social media company]Relevant, Negative

Source:  Capgemini Analysis

In a recent article published on dandodiray.com entitled Analyzing Securities Class Actions by Size7, the
author shows that the probability of a public company experiencing a SCA is correlated with its industry
sector and market cap. Using publicly available SCA court filings from 2011 through 2020, the author’s
statistical analysis suggests that the size of a company is positively correlated with the probability of
experiencing SCA. In other words, the larger a company is, the more likely they are to be named in a
class action lawsuit. Further, the author shows that in addition to the statistically significant trend for
size, the same probability is also positively correlated company’s industry sector. For example, given a
similar market cap, it is more likely that a company in pharmaceutical industry experiences SCA than a
manufacturing company.

This analysis validates current industry best practices in which D&O pricing models rely on market
capitalization and industry sector, among other variables, to price D&O policies. In other words, these
variables are deemed reliable to be used at scale for pricing purposes.

It is desirable for underwriters to access reliable analytics related to ESG given its increasing impact on
D&O losses. With existing ESG scores not generated based on a commonly accepted methodology, in
our analysis we developed and trained a predictive model that relies on the output of the NLP capability
presented earlier. In the next section we present the details of our analysis as well as its results
validating that the model can help underwriters increase their pricing and risk selection. Further, we
show that the generated segmentation is in addition to what underwriters currently use in their
processes, e.g., size and industry sector.

For a combination of public companies with and without SCA settlement losses, we analyzed a range of
publicly available documents using our NLP model. The purpose of the analysis was to (a) assess if text
extractions deemed as relevant to ESG risks are predictive, and (b) whether the predictive power of ESG
is above and beyond segmentation power underwriters get from variables such as company size and
industry sector.

To evaluate the predictive power of ESG insights, for companies with SCA settlement losses, we
analyzed unstructured data from those companies published prior to the plaintiffs filing an SCA. We
opted to perform our analysis on SCA settlements as the most relevant source of underwriting losses for
D&O policies. To compare the results of our analysis to those presented in the article entitled “Analyzing
Securities Class Actions by Size” we converted the results of that analysis, which was based on the
probability of an SCA, to the probability of an SCA settlement using the below equation:


Where P(SCA) is the probability of experiencing securities class action (SCA) and P(SCAS|SCA) is the
probability of experiencing SCAS conditioned on a company being subjected to SCA. The graph below
shows the variations of probability for a SCAS by company market capitalization followed by the table
listing P(SCAS) by industry and market cap.

Table. Probability of a SCAS by size and industry, 2015-2019

industrySmall CapMid CapLarge Cap
Finance and Insurance1.19%2.00%2.83%
Professional, Scientific and Technical Services1.22%2.04%2.89%
Education and Healthcare1.68%2.70%3.72%
All other industries0.84%1.45%2.13%

Source:  Capgemini Analysis

As part of our analysis, we generate an index called, αESG, which is weighted and aggregated based on
the ESG insights extracted by our NLP engine. The αESG index segments our sample to companies with
high ESG risks and low ESG risks. Our analysis shows that αESG is highly predictive and can predict
companies with more likelihood of experiencing a SCAS as well as those that are less likely irrespective
of their market capitalization. As shown on each graph, companies identified as “High ESG Risk” are
significantly more likely to experience a SCAS when compared to the case where ESG Risk is unknown,
the solid lines vs. the dashed line. Similarly, companies identified as “Low ESG Risk” are significantly less
likely to experience a SCAS when compared to an unknown case.

With respect to forecasting P(SCAS) by industry sector, we see that our ESG indicator can forecast
companies with a high and low likelihood of SCAS as shown in the table below.

To assess if the ESG predictive power of our index is in addition to the predictive power from company
market cap and industry sector, we conducted two correlation analyses. The first correlation analysis
was between our ESG index, αESG, and market cap of the companies considered. As shown in the below
graph by R2 metric, there is no correlation between αESG and market capitalization.

In the second correlation analysis, we looked at the correlation between αESG and industry sector for
small, medium, and large market caps. As shown in the graphs below, the R2 metric is low, indicating
there is no correlation between our ESG index and industry sector.

With ongoing discussions around ESG risks and their impact on D&O underwriting, we believe
transparent analytical capabilities can help underwriters and insurance companies to incorporate ESG
insights in their decision-making process in line with their long-term business strategies. Our analysis
suggests that investing in extracting verifiable ESG insights at scale is valuable.

The suggested analysis is also repeatable at low cost by design positioning the underwriting community
to rely on it as a powerful tool aligned with evolving definitions of ESG Risks.


1 Corresponding author: hesaam.aslani@capgemini.com

2 https://www.sec.gov/news/press-release/2022-46

3 https://www.loc.gov/item/global-legal-monitor/2021-08-17/germany-new-law-obligates-companies-to-establishdue-diligence-procedures-in-global-supply-chains-to-safeguard-human-rights-and-the-environment

4 https://plusweb.org/

5 https://www.businessinsurance.com/article/00010101/NEWS06/912347452/ESG-considerations-increasinglyfactor-into-D&O-underwriting

6 Aggregate Confusion: The Divergence of ESG Ratings, Review of Finance, Volume 26, Issue 6, November 2022, Pages 1315–1344, https://doi.org/10.1093/rof/rfac033

7 Analyzing class actions by size and industry sector https://www.dandodiary.com/2021/06/articles/securities-litigation/guest-post-analyzing-securities-class-actions-by-size/

Meet the Authors

Hesaam Aslani
Hesaam is a global AI leader as part of Capgemini Financial Services Insights and Data focusing on Insurance AI solutions and services. He has 18+ years of advanced analytics experience supporting global insurance and reinsurance organizations including QBE, Munich Re, and American International Group (AIG). Hesaam has 7 years of experience building property catastrophe models at RMS and a Ph.D., degree in structural engineering from Stanford University focused on building probabilistic loss estimation models for buildings.
Prior to joining Capgemini, Hesaam was the Senior Vice President of Analytics in QBE North America reporting to the Chief Underwriting Officer. He delivered pricing, claims and portfolio analytics for the QBE North America businesses, namely, Commercial and Specialty, Alternative Markets, and Crop. In his role, Hesaam delivered innovative pricing models across property and casualty lines including public company D&O, private company D&O, EPLI, General Liability, and other Specialty, Commercial and Personal Lines.
Before joining QBE, Hesaam was the head of Analytics, Strategy and Accumulation at Munich Re America. He was responsible for the economic capital modeling and reporting to the CFO, as well as development of pricing models to support business growth and to enhance risk selection.
Prior to joining QBE, Hesaam was the Sr. Director of Insurance Analytics at AIG, responsible for model risk management for all models used by AIG General Insurance businesses including P&C pricing models, and capital models.
Mohammadreza Iman
Mohammadreza is a lead data scientist, AI/ML solution architect, and an active researcher in AI/ML. He has a PhD in computer science and is the author of multiple scientific publications in the AI/ML research area. He has over 20 years of extensive experience in IT and computer science in industry and academia.
For over seven years, he has focused on solving problems and researching data science, machine learning problems, and artificial intelligence. He is currently working as an AI/ML solution architect for one of the largest insurance companies, bringing out-of-the-box solutions to the industry.
LinkedIn; G-Scholar
Jonathan Zwi
A former professor whose academic engagements spanned the arts and the sciences, Jonathan Zwi has been the recipient of numerous honors and awards in the fields of music, visual arts, philosophy, education, and neuroscience.
After earning his doctoral degree in music, Jonathan formally transitioned into data science and software engineering by building an application to perform real-time extraction and analytics of iceberg orders executed on the order books of the NQ and ES on the Chicago Mercantile Exchange.
Jonathan remains deeply committed to staying at the forefront of advancements in AI/ML, in addition to serving in data science, developer, and machine learning engineering roles for industry-leading Financial Services organizations across both Banking & Capital Markets and Insurance.
He continues to enjoy getting his hands dirty writing code, as well as leading teams of software engineers and serving as a mentor to the next generation of AI/ML professionals.
Jason Preciado
Jason is a dedicated data scientist with a bachelor’s degree in computer science and a passion for learning. With a solid foundation in programming fundamentals, he has jumped into the world of data analysis and AI/ML soon after completing his undergraduate studies.
Over the past couple years, he has been actively engaged in various data-driven projects and makes use of machine learning techniques, large language models, and other generative AI solutions to extract valuable insights. He is excited to continue to learn about new data science and AI/ML technologies to make a meaningful impact in his field.
Cameron Raynie
Cameron is a developer and data scientist with a bachelor’s degree in computer science. Deeply interested in understanding the underlying methods through which the world works, his work in data science has been a rewarding experience.
Initially working as a developer on several personal projects, Cameron’s move into the field of data science has put the world of AI and ML into a new light. Excited to continue their journey into data and AI, Cameron is looking forward to developing more data driven AI and ML solutions.