SESAMm CMO Jorge Alvarez explores the importance of obtaining comprehensive and meaningful ESG data, and how machine learning is revolutionising the way investors can approach this key workflow.

Environmental, social and governance (ESG) data is a valuable tool that’s become a standard measurement in sustainable finance for corporate stakeholders. However, due to the growing demand and need for accurate and timely ESG data in investment decision-making and the ESG finance field, it’s also difficult to attain.

Besides implementing ESG principles and policies, companies are asked to provide information and reports on related performance in a consistent and standardised format. As one might notice in this scenario, ESG data comes primarily from the very companies we want to evaluate, creating somewhat of a conflict.

Today’s ESG data challenges

At their core, ESG metrics capture a company’s performance on a given ESG issue. When this aim is achieved, investors can use the data to evaluate and hold companies accountable for their ESG performance. But how would you know whether ESG data accurately captures a firm’s performance?

In the Journal of Applied Corporate Finance, Sakis Kotsantonis and George Serafeim share ‘Four Things No One Will Tell You About ESG Data’. Here’s a summary:

• ESG measuring, data and how companies report them are inconsistent
• Lack of benchmarking transparency undermines the reliability of peer performance ranking
• ESG data providers deal with ‘data gaps’ differently, and their gap-filling approaches could lead to significant discrepancies
• Interpretation differences among ESG data providers are considerable and are growing with the quantity of data becoming publicly available
• AI to meet rising ESG data demands

Even as investors consider ESG one of the many major market factors, sourcing and analysing data remains a problem. “The absence of standardised ESG datasets and reporting methodologies makes it difficult for issuers to disclose meaningful information on sustainability,” according to a post on WorldQuant.

But despite the data limitations, ESG investing demands continue to grow. For instance, in its 2021 Key Findings, RBC Global Asset Management found that 75% of respondents from 800-plus institutional investors had integrated ESG principles into their investment approach, an increase from 67% since 2017.

Machine learning helps with this demand. For instance, advances in natural language processing (NLP) in machine-learning techniques have made it possible to extract unstructured data from web sources, like news, blogs, forums and social media, to gain timely and accurate ESG insights. This alternative data has been integral for seeing an entity’s ESG controversies or events in near real time, providing a unique perspective on ESG data and details, filling the data gaps more accurately.

NLP algorithms can read billions of news pieces, articles and text-based web data. They categorise extracted data and can determine positive and negative sentiments, producing potential predictive indicators. Investors and researchers can use NLP to mine keywords and categories of underlying data, to evaluate portfolio companies or see their exposure to ESG factors.

Some ESG rating agencies are now integrating or outsourcing NLP-derived datasets into their processes to extrapolate ESG scores. Likewise, investment firms, like asset managers, are incorporating NLP-enhanced web data into risk management, especially when looking into private equity-type assets. Many are meeting their needs with NLP companies, such as SESAMm and others.

How SESAMm extracts ESG data

SESAMm has one of the largest data collection sources to extract data from (known as its data lake), and its NLP machine-learning algorithms are tuned specially to key indicators.

SESAMm’s data lake is unique and ideal for investment research and advanced analytics, given its depth and breadth, the fact that it includes more than 100 languages, and that it is updated in near real time.

Including data since 2008, the data lake consists of more than four million sources made up of more than 20 billion articles, forums and messages, such as professional news sites, blogs and social media, increasing by an average of six million per day. The data lake is also updated hourly to give investors near real-time insights into their investment interests.

Moreover, the coverage is global, with 40% of the sources in English (US and international) and 60% in multiple languages, including Japanese, Chinese and Eastern European. We select and curate these sources to maximise coverage of both public and private companies, focusing on quality, quantity and frequency to ensure a consistently high input value.

Furthermore, SESAMm’s developers tune the machine-learning algorithms for key indicators such as mention volume, sentiment analysis and emotion, ESG and SDG. Additionally, they optimise the structure and schema for optimised SQL queries.

For example, our knowledge graph, a digital representation of a network of real-world entities, puts the schema in context through semantic metadata and linking, providing a framework for analytics, data integration, sharing and unification. In other words, we map and label the concepts, entities and events, and connect and identify their relationships for quick and accurate recall.