The breadth and depth of Environmental, Social, and Governance data that investors currently have at their disposal is “a good problem to have”, as it highlights the normative sea-change over the past decade towards a now crucial commitment that companies must make to ESG transparency. Machine learning can help in discerning – and when necessary, creating - signal from noise.
Making sense of the massive amounts of ESG inputs with machine intelligence systems
The adage “a good problem to have” is apropos when looking at the breadth and depth of Environmental, Social, and Governance data that investors currently have at their disposal. Even a single company can have thousands of ESG-related data points, let alone entire industries. That’s the good part because it speaks to the normative sea-change over the past decade towards a now crucial commitment that companies must make to ESG transparency. The problem part is trying to make sense of it all such that ESG transparency amounts to real transparency. Here is where advanced machine learning techniques can have a big impact.
Advancements in the suite of novel tools and technologies we collectively call “machine intelligence” bring several key benefits to the world of ESG investing, including the ability to:
- Rapidly process the barrage of new ESG data sources that are now available
- Systematically apply the rigor of fundamental analysis to ESG investing
- Go beyond headline ESG ratings to identify materiality across industry, geography, and time
- Look through attempts by management to window-dress or “greenwash”
- Customize specific ESG solutions, such as optimizing for lower carbon while holding the pursuit of alpha constant
Voya IM’s Equity Machine Intelligence (EMI) team has long focused on exploiting these benefits to unlock value from the full mosaic of company fundamentals, sentiment and ESG data.
Discerning – and when necessary, creating - signal from noise: The next frontier of ESG investing
Not only can machine learning systems more quickly process mountains of corporate ESG data, they can do so more effectively. This means that ESG integration can go well beyond the well-known limitations of headline ratings.
Machine learning techniques such as natural language processing (NLP) can be used to gauge the sentiment of news stories, and thus tap into changing social or community perceptions of a company.
This capability is becoming a crucial one, as ESG data providers proliferate. In “ESG Ratings and Rankings: All over the Map. What Does it Mean?”1 it was noted that there are over 150 organizations offering upwards of 600 ESG data and ratings products. This has fed another problem — the lack of standardization of ESG data — a difficult issue for traditional quantitative tools to address. There are also significant methodological divergences between ratings provided by various ESG providers.2 Indeed, the average correlation of ESG ratings between different providers is only 0.54.3
By digging deep through both structured and unstructured data, machine intelligence systems can unearth signals from the noise faster, and in many cases detect signals that may have been previously unknown. These systems can also impute missing values — or fill the holes — in data sets, bringing new clarity to messy data. The goal is to integrate the best of what ESG data has to offer with the mosaic of other fundamental and sentiment data to help discover future stock outperformers.
This approach can help identify opportunities that could be easier to overlook than one might imagine. For instance, Hormel Foods Corporation (ticker: HRL) was identified in January 2013 by EMI models as a potential long-term outperformer. At the time, while Hormel Foods’ ESG rating was modest according to published MSCI ESG data, the model identified the company as underappreciated and saw that key fundamental and ESG metrics were beginning to improve. The model was right; over time the company made substantial improvements and became an ESG leader (Figure 1). During the more than seven-year holding period, the company’s market cap doubled from around $10 billion to over $20 billion, and the stock outperformed the Russell 1000 index by nearly 90%.
Shining a light on greenwashing opacity
With this bottom-up, fundamentally anchored perspective the EMI team has been unlocking value from the deep analysis of ESG data for almost a decade. The underlying philosophy has remained consistent: to treat ESG data as a key “alternative” data source, and to use it with the same rigor of financial analysis; to unearth risks and opportunities in the same way that the system forensically analyzes company financials.
Consumers and investors alike are pressuring companies to demonstrate their ESG credentials, either through actions at the corporate level, including in the products they offer, or via the instruments they use for financing. Governments and regulators have also introduced enhanced standards and regulations to support increased disclosure and consistency of ESG reporting. This ESG push, which has been propelled by legitimate risks and concerns important to market participants, has led to an inevitable rise in the volume of “green” claims made by companies attempting to demonstrate sustainability credentials to their stakeholder base (Figure 2). However, the sheer volume of ESG marketing and labelling, in combination with nonuniform sustainability commitments and reporting, has made it increasingly difficult for stakeholders to identify which claims are trustworthy and reliable, and which are not (i.e., “greenwashed,” in industry parlance).
Source: FactSet. Copyright © 2021 by Standard & Poor’s Financial Services LLC. All rights reserved.
Indeed, at the recent September Skybridge Alternatives (“SALT”) conference in New York, Gareth Shepherd, portfolio manager and co-head of the EMI Team, noted that some corporate treasurers and CFOs have figured out that if they can “window dress” their companies’ ESG characteristics, they will get increased investor flows, as well as higher marks from rating agencies. Such perverse incentives are why it may not be prudent to rely on just headline ratings.
Case in point: FACEBOOK
On March 17, 2018, The New York Times published an article that Cambridge Analytica used “private information from the Facebook profiles of more than 50 million users without their permission… making it one of the largest data leaks in the social network’s history”.4
Prior to the scandal, Facebook’s ESG ratings5 had been improving across several different providers. After the news broke, Facebook’s stock fell more than 20% over the following two weeks from a peak earlier that year. At one point later that year, the stock was down more than 35% as the scandal widened. Ratings agencies reacted by downgrading Facebook’s ESG score6… a little too late.
Fortunately, the EMI team’s machine learning models flagged Facebook on ESG issues before the scandal came to public light, placing it in the bottom decile of companies in the U.S. universe. We believe this to be a testament to the machines' ability to go beyond headline ratings and forensically dissect underlying data without emotion or human biases.
Nowhere to run, few places left to hide
With the application of machine intelligence, there are few places left to hide. Companies tempted to prop up their ESG performance for ratings agencies and stakeholders alike are forewarned: ultimately, they will be discovered if the data does not match the rhetoric.
In our next blog, we will explore further the practical side of integrated ESG investing and how specific signals, such as the momentum of ESG, can enhance stock selection.