ESG Anomaly Detection
LSEG Labs Project: ESG Anomaly Detection
A tool to identify anomalies in environmental, social and governance (ESG) data
Trust and transparency issues are a common pain point in data related to ESG (environmental, social, governance) - there is an abundance of disclosure and ratings data from various sources, yet it is often not entirely reliable, consistent, and comparable. This increases not only time and costs of users in researching and processing ESG data, but also makes it a challenge to take fully informed decisions and actions.
Common reasons why an anomaly may occur
(the list is not exhaustive):
- Inaccurate data due to unintentional errors
- First time reporting
- Significant changes within a company motivated by strategy or operations
- Structural changes within an industry group
We harness the power of AI (Artificial Intelligence) and data analytics to derive insights from sources of ESG data. Leveraging Machine Learning (ML) and statistical analysis, we are developing anomaly detection capabilities to highlight significant trends of ESG scores and the underlying components.
This includes anomalies that occur temporally as well as across peer companies in the same industry group. Users are then able to adjust their portfolios based on the anomaly intelligence gleaned. In this way, we enable regulators, asset managers, advisory and corporates to process high volume ESG data across multiple dimensions and timeframes to act on insights efficiently.
The anomaly detection is divided into five steps:
- ESG category score changes are calculated with respect to previous years
- Z-scores of the changes are calculated over each industry group
- A machine learning model is trained with data over 18 years and around 10,000 companies to detect outliers
- A model explanation tool is applied to quantify how much each ESG category contributed to an anomaly
- The model outputs are mapped to a magnitude and polarity score to make them easily interpretable
The panel above visualises the result of the anomaly detection. Each dot represents a stock of a portfolio. The anomaly magnitude on the x-axis reflects how unusual the ESG scores of a company are: the higher this value the more unusual. The anomaly polarity on the y-axis explains the trend of the ESG scores: the more positive this value the more the scores have increased compared to previous years.
Delivery mechanisms - this flexible solution could reside on:
- Interfaces on Refinitiv Workspace / Eikon
- Data feeds / APIs that could be served directly to users
The Path Forward
The team is continuing to explore the underlying factors contributing to anomalies and combine these is with insights into emergent trends from news and other sources, enabling users to deepen their confidence in the quality of their portfolios / holdings.