Home > Research > Publications & Outputs > ControversyBERT
View graph of relations

ControversyBERT: Detecting Social Controversies and their Impact on Stock Returns

Research output: Contribution to Journal/MagazineJournal articlepeer-review

E-pub ahead of print
<mark>Journal publication date</mark>26/07/2023
<mark>Journal</mark>Journal of Impact & ESG Investing
Publication StatusE-pub ahead of print
Early online date26/07/23
<mark>Original language</mark>English


Companies linked to social controversies experience an average drop in returns of more than 200 basis points in the days surrounding the outbreak of controversial news. To identify such social controversy events we build ControversyBERT, a large language model trained on a sample of 1 million news headlines to detect reports of controversial incidents in daily news feeds. Among the eight examined social dimensions, controversies surrounding violations of product safety standards, labour standards, as well as consumer data safety and data privacy breaches significantly affect firm returns. The corresponding stock price reaction is negative in all considered geographic regions and is largely driven by small to medium market capitalization companies for which information diffusion is slowest. Even though the build-up in controversy news sees most of the negative price reaction occurring before the event, our controversy indicator can help avoiding about 30% of the overall effect by the timely divesting holdings in the identified companies.