Subscribe
Featured Image

Detecting Statistically Anomalous ESG Events in Real Time with AI

ESG

By Shirley Birman  |  June 3, 2021

Environmental, social, and governance (ESG) topics frequently manifest themselves in the form of positive or negative events that can significantly and suddenly impact the value of a company. One of the challenges in ESG analysis is determining the level of significance of an ESG event and how material it is to a company. Material events impact company performance in different ways; surprise or abnormal events are likely to result in stronger price responses.

In a recently released research paper, Detecting Statistical ESG Anomalies, the FactSet ESG Solutions and Research team discusses how to identify the most statistically significant Spotlight Events across a portfolio for a given time range using artificial intelligence (AI). The goal of the study is to help identify the most salient and relevant spotlights and provide a systematic approach to surface these events for both fundamental and quantitative use cases. This can help investors eliminate ESG blind spots, better understand emerging material issues, and develop new systematic strategies that incorporate timely ESG events data.

Detecting Anomalous ESG Events with Information Flow and Score Changes

Spotlights are a beneficial solution to alert investors to impactful developments in a company’s ESG performance. Using the historical, point-in-time spotlight data, investors can narrow down the spotlights to truly capture those with abnormal activity (i.e., high-impact ESG events). For each spotlight, a z-score is calculated based on volume of information flow combined with ESG score change on the trigger date of the event.

Our research paper confirms that spotlights with higher z-scores show higher absolute spread returns from one to 20 days post trigger date, indicating the importance of detecting ESG events in real time. The figure below plots the mean quintile spread where the universe is bucketed by the spotlight z-scores. Companies with high z-score events can be identified as having either an abnormal amount of article volume, a significant change in ESG score, or a combination of both. The Spotlights historical data allows for straightforward calculation and detection of these anomalous spotlights.

Spotlight Event Analysis

Since Truvalue Labs tracks volume of ESG relevant articles for companies in real time, we can build a signal that enables the detection of an abnormal amount of information flow. Other ESG solutions do not track ESG event volume historically, making the development of such a signal impossible.

Surprise vs. Reinforcing Spotlight Events

FactSet Truvalue Labs Spotlight Events provide the opportunity to refine events based on their expected significance as determined by the volume of articles published on that event relative to the typical ESG news flow for that company. Spotlight Events can further be refined as a function of their ESG score change. Specifically, the events can be distinguished as surprise or reinforcing. For example, a sudden and unexpected data breach is an excellent example of a surprise ESG event. On the other hand, some oil companies have been sued for carbon emissions for years—and both new lawsuits and new phases of existing lawsuits are not surprising—but rather reinforcements of what has already been discussed heavily.

There is motivation to think about the expected differences between surprise and reinforcement events. Surprise events lead to a strong change in ESG score on a particular topic, while other events serve more as reinforcements that increase the confidence that the current ESG score on the topic is correctly measuring positive or negative sentiment.

The figure below shows the quintile spread of both reinforcement events (light blue bars) and surprise events (dark blue bars) and on a global universe of companies from 2016 through June 2020. The volume threshold is determined by the article flow typical for the company over the past 12 months. Spotlights are created using a dynamic volume threshold that updates for each company daily, where the thresholds are set based on the levels of a company’s trailing 12-month article volume on a given day. For example, Volume >= 1 represents all spotlights in the global dataset, whereas Volume >= 2 incorporates only those with at least two times the information flow relative to the threshold for the company. The results show that events with at least two times the volume threshold maximize returns and are particularly useful for further examination in shorter-term strategies.

Global Spotlights Events

Using Statistics and AI to Prevent Information Loss

Most conventional approaches to ESG event detection require an ESG analyst to label an event as low, medium, or high impact. With such an approach, the end user loses the ability to perform meaningful statistical analysis such as the z-score and volume threshold approaches discussed in this write-up. For example, how would one compare all “high-impact” spotlights among one another in a conventional approach to ESG event detection? This mathematical information loss is a critical deficiency in conventional approaches.

Using Truvalue’s AI methodology ensures consistency and timeliness in identifying anomalous events. It also allows for ranking events versus historical volumes of information flow and ESG score changes to capture the most impactful events. Events with abnormal article volume or large changes in ESG scores merit special attention. FactSet Truvalue Labs Spotlight Events data truly provides investors with a unique offering to make informed investment decisions to critical ESG events.

detecting-statistical-esg-anomalies

Shirley Birman

Quantitative Researcher, Truvalue Labs

Ms. Shirley Birman is a Quantitative Researcher at Truvalue Labs, a FactSet company. In this role, she collaborates with the sales team to identify opportunities and serves as a subject matter expert on the various analytics provided by Truvalue Labs. She works with both quantitative and fundamental clients, highlighting how they can integrate Truvalue Labs’ data and analytics into their investment process. Prior to joining Truvalue Labs, she was Manager of Quantitative Consulting at Thomson Reuters/Refinitiv, focusing on the expansion of StarMine analytics and models across both quantitative and fundamental clients. She also worked at StarMine (acquired by Thomson Reuters) as a quantitative research analyst. Ms. Birman earned a master’s degree in finance from The University of Texas at Austin and holds a B.S. in systems engineering from Washington University in St. Louis.

Comments