This article is a part of NordSIP Insights – Handbook – “Systematically Sustainable”.
by Richard Tyszkiewicz
We are often reminded that sustainability begins at home with the actions and changing habits of individuals. While this is true, it is also likely to be too little, too late unless the trillions worth of global institutional investment money follows suit. It is encouraging to see packed-out responsible investment conferences in recent years, as well as concerted efforts by industry and regulators to standardise Environment, Social and Governance (ESG) investment practices and metrics. It’s still a work-in-progress, but pioneering asset owners and managers are continuing to deepen their commitments to sustainability across asset classes. Meanwhile many “ESG laggards” are struggling hard to catch up, having been slow to realise that this is no longer an ethically driven side-show. Sustainability has become a compelling priority from risk/return, fiduciary and commercial perspectives. However, an industry built on the analysis of financial data is now crying out for reliable, comprehensive and timely sustainability data.
Specialised ESG ratings agencies were first established in the 1980s, catering mainly to ethically or environmentally driven investors looking to implement negative screening on their equity portfolios. Many of these early innovators were acquired during the last decade by larger ratings firms such as Moody’s, Morningstar and MSCI. They continue to develop their methodologies and provide much needed input to investors, but the general reliance on ESG information published by the scrutinised companies themselves can be problematic.
In many markets mandatory reporting is still limited to Greenhouse Gas (GHG) output or carbon footprint data, for instance. Scope 3 numbers, or the environmental impact of full product lifecycles are often not available. Resulting gaps in the data may have to be plugged by imputed or assumed data. ESG ratings firms must also deal with the inconsistent availability of social metrics as well as the overall positive bias of companies’ self-reporting.
These challenges, compounded by differences in proprietary ESG factor weightings, lead to significant inconsistencies in the final ratings. An MIT Sloan School Working Paper published in August 2019 revealed an average ESG rating correlation of only 0.61 between five well known agencies, which contrasts with near perfect correlation among credit rating agencies. Finally, company reporting cycles mean that ESG information can be somewhat static and out-of-date.
Regulators, Central Banks, supranationals, NGOs, industry lobby groups and ESG ratings firms themselves are all working hard to improve the quality and flow of non-financial company data. In the meantime, some innovators have instead been looking to exploit alternative data sources. The basic concept is not new, with old stories of hedge fund managers getting employees with clickers to manually count cars in retail parking lots as an alternative indicator of company sales projections.
What has radically changed is the explosion of available data on the internet. This Big Data pool would be impracticable to trawl through manually, but vast computing power can now be obtained relatively cheaply, which opens up the possibility of using Artificial Intelligence (AI) and Machine Learning (ML) to glean new sources of ESG data. This approach is used by firms such as San Francisco based TruValue Labs, which processes over a million data points each month from around 100,000 on-line sources. Efforts are made to curate these sources, for example by seeking to identify academics and industry experts and weed out private individuals. This improves the overall quality of the data, which is put through proprietary algorithms designed to identify positive and negative ESG related information. ML allows improved accuracy over time as recorded items are cross checked with historical data and short-term impacts smoothed out. By drawing from a multitude of independent sources this type of analysis effectively addresses the potential bias of company self-reporting.
Moreover, while traditional ESG analysis has been well established in developed equity markets, AI based firms such as TruValue Labs, Sensefolio and Arabesque S-Ray are able to cover a much broader spectrum of global markets and asset classes. The only limitations are language coverage, processing power and the ability to design algorithms that can correctly detect and interpret relevant positive and negative ESG information. The aforementioned MIT study stated that the discrepancies were partly caused by different weightings assigned by each firm to the various E, S and G criteria. In the absence of a single universally agreed framework, users of these AI-based tools can fine tune the output to suit their chosen criteria, be they proprietary or based on standards such as SASB, GRI or the UN SDGs.
Asset managers are also using these new data feeds to try to identify new sources of alpha. Successful ESG managers typically exploit all the internal and external data sources they can find, while fully understanding their respective limitations and differences.
The information dislocations that are hopefully being addressed by standardisation initiatives can be a source of investment opportunities for these managers in the meantime. A poorly rated company might be on an upward trajectory in ESG terms, but still under the ratings radar. Some complex, multi-faceted ESG issues can also be missed. For instance, aluminium can manufacturers are typically penalised for energy use, but their products benefit from far higher recycling rates than plastic, which is now the focus of globally negative sustainability headlines.
Rather than producing ESG ratings, AI algorithms can also be designed to identify correlations between available ESG scores and actual market data. Such methods will potentially put to bed the argument that sustainability comes at the cost of portfolio returns.
While some of these new ratings providers have already shown what can be drawn from large numbers of properly curated news and commentary sources, the range of data sources being made available is constantly expanding. It is now possible for investors to buy data feeds from satellites, drones, the so-called “Internet of Things” and general web traffic, which is prompting any number of interesting and creative approaches.
For instance, one might try to verify officially stated oil reserves using the size of the actual shadows cast by oil drum stacks. Activities such as deforestation and open cast mining can also be accurately investigated. UN SDG #6 focuses on clean water and sanitation, and many investors are realising that water scarcity is a rapidly increasing risk for agriculture, industry and society at large.
Montreal based Aquantix is a good example of firms seeking to mine this new world of data. They use AI to collect around 20 million water related data points each day, making the resulting analysis available to investors looking to build portfolios that positively address that specific SDG while mitigating long-term portfolio risk.
The world of sustainable investing can seem rather fast-moving and often ill-defined to institutional investors trying to keep abreast of developments. Efforts by the likes of the EU to set a common language, or taxonomy in place to help avoid confusion and greenwashing are to be applauded and will undoubtedly help as they bed in. Coupled with the innovative use of new data sources and AI, this should help investors position their portfolios on the right path to achieving sustainability goals and avoiding risk as the global economy shifts away from long-established detrimental patterns of production and consumption. So far, the AI based ESG ratings tend to be used as a complement to the more traditional research-based agencies’ output.
Given the complexity and variety of sustainability themes faced by investors, it’s generally accepted that expert human interpretation of data is a vital element in the process. The possibilities presented by AI and Big Data are nevertheless fascinating and a very positive development. It will hopefully speed up efforts to shine a light on areas of the market such as fixed income and bank lending that currently largely evade the ESG radar.
Thanks to Machine Learning, one can also expect greater overall quality of output over time. There is some concern among academics that commercial imperatives and the need to amortise the cost of gathering certain specialist data might inhibit its positive application from a global sustainability standpoint. Nonetheless, it is clear that Alternative Data and its exploitation via AI can be expected to gradually shift into the mainstream as large investors, index providers and regulators start to integrate it into their products and processes.
Featured image ©Shutterstock
