But questions of whether and how we can actually improve investment performance with ESG-related information have been obscured by the proliferation of uninformative and misleading content and claims.
The inconvenient truth is that readily available ESG data, including off-the-shelf ESG ratings, is unlikely to be of investing value for investors seeking to maximise risk-adjusted returns. It would be surprising if it were. There are a number of reasons for this.
Subjectivism
ESG ratings are not created (solely) for financial forecasting purposes, and they reflect the purveyor’s business incentives and all kinds of judgment calls. It’s no surprise, therefore, that there is enormous dispersion across ESG ratings, reflective of deep disagreement over what concepts to evaluate, how to measure those concepts, and how to weigh the metrics. For example, a fossil fuel company might outscore an energy transition company depending on the relative weighting of environmental versus social and governance criteria.
Contrary to Elon Musk’s recent assertion that ESG is a scam, however, ambiguity in ratings hardly implies that there is no value in ESG.
Data issues
It’s no coincidence that ESG has developed concurrently with a revolution in systematic investing – the application of artificial intelligence (AI) to analyse alternative data, i.e., information that was originally generated for non-financial purposes. That’s because there is enormous overlap between ESG-related information and the types of alternative data now being exploited by systematic investors, examples of which would include free-form text in corporate communications, regulatory filings, and media reports. ESG-related alternative data is often unstructured, ungoverned, and ‘big’.
While those characteristics make the information difficult to work with, that challenge represents the seed of opportunity for systematic investors.
The value of systematic application
The key to generating additional value from ESG lies in the application of a sophisticated toolkit to access, process, and analyse messy alternative data. Such methods include:
a. Information retrieval – automated processing of free-form text and other alternative data sources to extract information that is material to investing. A prototypical example in the ESG context involves mining regulatory filings for significant disclosures about ESG-related risks. Doing so would be infeasible without systematic machinery (or an army of analysts) given the dauntingly long and tedious nature of such reports.
b. Unsupervised machine learning – a class of AI algorithms that automatically identify patterns in data.
c. Supervised learning – the application of machine learning techniques in combination with human judgement, often to improve forecasting or to make subjective judgments.
We believe that the systematic investing process is uniquely well-suited to achieve ESG objectives. The application of AI to analyse the alternative data that is so central to ESG represents the natural evolution of the systematic toolkit. It has perhaps surprisingly broad relevance to ESG, across alpha generation, risk mitigation, engagement, and in aligning asset owners’ portfolios with their values.
Andy Moniz and Seth Weingram, Acadian Asset Management LLC
