How data science is being used to advance the measurement and assessment of ESG performance.
1. Current issues with ESG data
The need for businesses to act with the environment, society and governance practices in mind to make material improvements to the financial, legal and reputational performance of the firm, and in some cases, to create a positive impact, is coming to light. This realisation is driving investors and financial institutions to begin integrating such ESG considerations into their reporting and risk management frameworks.
The reporting of ESG performance by companies has progressed over the last couple of decades, but it is still in its early stages. It is neither obligatory nor standardised, unlike financial reporting. There are few regulations, especially in the US, that require ESG disclosure, and very few legal repercussions for violating them. This means that, despite the increasing availability of ESG data, often the information that companies provide can be unreliable or incomplete. For example, self-reported ESG information can lead to greenwashing, as companies try to appear environmentally conscious, but are not taking any notable action to become more sustainable.
2. Addressing the flaws of ESG data with data science
The role of data in measuring ESG performance
Artificial Intelligence and data science can be used to fill in the current gaps of ESG reporting.
Data Science can improve the reliability of reported data.
The Corporate Sustainability Reporting Directive (CSRD) plans to mandate data on sustainability performance in company annual reports with third-party auditing by 2025. Until then, reported data remains inconsistent and unreliable. This applies to even one of the most widely used and financially material environmental metrics, Scope 1 CO2 emissions, where there is a high variability among data sources from errors, lack of standardisation and poor data quality.
By using data science, in the form of machine learning algorithms and other techniques, very large data sources can be utilised to detect outliers, automatically select the best data source for overlapping data, and to accurately estimate non-reported data to ensure companies are being transparent in a quick manner without the additional costs of external audits.
Machine Learning (ML) can expand data coverage.
It has been found that on average, US SEC-registered firms provided only about 18% of the disclosure items that the Sustainability Accounting Standards Boards (SASB) considers material to the financial performance of the company. Given that ESG reporting is not mandatory, often companies only report data if it is favourable to them. For example, reporting on employee ethnicity has gone up, but these are likely to be companies with a ‘best in class’ diverse workforce. The inconsistency of reporting then makes it very difficult to consistently score companies on their performance, and potentially puts companies that purposely leave out information ahead of the transparent ones.
Machine learning algorithms can be used to fill in these gaps to give the most complete version of a company’s ESG performance. ML algorithms use data inputs to build a model that makes predictions for how to deal with the new data. For example, Clarity AI’s ‘estimation models’ use ML to estimate sustainability metrics by deriving them from other already reported, easier to measure features about a business. For instance, the model uses information about a company’s industry, its production process, and its labour costs to estimate the level of CO2 emissions it is generating.
Natural Language Processing (NLP) can inform how sustainability metrics develop.
There are several limitations of the manual assessment of news to assess companies’ behaviour. Firstly, the assessment of data is subjective and perhaps biased; there is also a limitation in the volume of data that can be covered by a person; and there is a limit to the amount and quality of insights that can be generated without technical analysis. This makes it harder to provide up-to-date and meaningful information in a consistent, transparent way.
NLP algorithms use data inputs to build a model that extracts information from text. In this case, these data inputs consist of a large number of articles covering different ESG factors to allow the model to learn the relevant criteria to extract the relevant information from new articles. It can be used to automatically:
Detect news articles that contain relevant ESG information.
Classify the information under the appropriate type of ESG factor (e.g. biodiversity hazard, worker conditions).
Judge whether the impact is positive or negative, and the size of the impact.
The model enables more data to be covered at a quicker rate, and can be updated frequently as more data becomes available whether in the form of reports or news articles. Clarity AI’s whitepaper on how data science can enable SFDR reporting goes into greater technical detail on how exactly models are created and tested for accuracy.
3. The Limitations and Potential of AI and Data Science for ESG
From a technical standpoint, there are clear benefits to using AI and data science methods to inform the assessment of a company’s ESG performance. However, although investors value diverse sources of information for verification purposes, whether this extra information is generating enough additional value is weighed against the costs of processing it.
Using ESG data and analytics providers can be costly in terms of subscription fees, analyst time to go through the insights, and potentially the costs of making wrong decisions based on incorrectly or poorly processed data. Automated information processing can, on one hand, reduce how long analysts have to work (and so lower human resource costs). However, on the other side, it may increase the complexity of analysis and model risk if it reduces the likelihood of incorrect data or faulty inferences being discovered. For investors that do not need to consider how ethical their investments are, the cost of integrating these ESG data analytics need to correspond to an increase in expected financial returns.
ESG reporting has come a long way, but still has a long way to go, especially in terms of data limitations. AI and data science can be one of the fields that help advance the measurement and assessment of ESG, with further potential in other applications of ESG, from driving boardroom conversations using ESG data-driven insights to even tying ESG data to compensation and bonus schemes.
Comments