Millennials demand social impact from investee companies. Can AI be applied for better ESG estimation?
Environment, Social and Governance (ESG) investing is a process where investors consider not only the financial factors but also ESG factors before making an investment decision. While ethical considerations in deciding investments is not new, the adoption has accelerated lately. In fact, in Europe 58.8 per cent of invested assets are run according to an ethical mandate. Moreover, it is projected that millennials could put between $15 trillion – $20 trillion into U.S.-domiciled ESG investments as they demand not just financial returns but also social impact from the investee companies.
Considerable debate centres around the contradictions in impact investing where at times it maybe better to invest in companies in the “dirty fuel” (coal, oil) industry than in ones which flout corruption or child labour norms.
by Mrinal Mishra.
Due to the aforementioned shifting trends, it becomes imperative that wealth managers, investment analysts and family offices are able to use public (and private) information to determine the ESG rating of prospective investments. However, this process is not costless and involves significant investment in terms of time and money to be able to deliver the best possible results. While ratings and self-reported measures (by companies themselves) of ESG do exist, they are not sufficient by themself. This is because a vast wealth of information which could impact ESG rating of companies is in the form of unstructured data (primarily text, images). It is in this context that AI can solve or attempt to make the problem easier to tackle for prospective stakeholders.
AI can make the ESG problem easier to cope with due to its ability to generate better insights when given large amounts of data. Specifically, this can be done in two different ways. First, AI can deal with bigger quantities. A large amount of high frequency news flow, tweets and public reactions can be mapped to be used to generate insights which might be beneficial to the end user. Second, it can deliver better quality. Existing high quality data sources like annual reports and conference calls can be analyzed for tenor and orientation of speech using NLP techniques. This enables one to generate insights which are not evident immediately and require careful reading and analysis to arrive at. AI solutions which solve either of the two or both problems simultaneously can open the doors for a more scientific approach to ESG investing.
However, challenges to applying AI effectively for better ESG estimation remain. Most of the textual information in the public domain is put out by the companies themselves and thus tends to have a positive bias. Designing cutting edge models cannot be a substitute for poor input data. Hence, even if one trains a well-crafted model on incorrect data, it will ultimately produce erroneous results. While bodies like the Sustainability Accounting Standards Board (SASB) are looking to address this conundrum, the frequency of reports is also a concern. Most assessments and disclosures are published annually (or at most quarterly). This calls for an “outside-in” approach to ESG data acquisition whereby insights should be arrived upon by gathering news and then using it for AI models to deliver high calibre results. Finally, to design supervised models one needs a precise estimate of the universally accepted truth. However, this is not the case yet for ESG factors. There seems to be limited clarity on which variables effectively benchmark the ESG rating of a company thus limiting the predictive ability of supervised learning models. However, on the flip side it presents an opportunity for one to employ unsupervised learning models (supervised learning estimates parameters based on past outcomes where as no such outcomes are required for unsupervised learning) which address the drawbacks faced by supervised models.