CMStatistics 2022: Start Registration
View Submission - CFE
Title: When firms open up: Identifying value relevant textual disclosure using simBERT Authors:  Christian Breitung - Technical University of Munich (Germany) [presenting]
Sebastian Mueller - Technical University of Munich (Germany)
Abstract: By introducing simBERT, a novel semantically sensitive similarity measure for textual data, we find that international annual reports contain value-relevant information that investors do not timely price. We measure the value relevance of international corporate disclosures by constructing a portfolio that is long in stocks with a low- and short in stocks with a high level of semantically new information. Such a portfolio yields a highly significant yearly abnormal return of 8.52\%. We observe a higher value relevance of textual disclosure in developed countries, which we trace back to stricter securities laws standards. Our findings thus indicate that tighter regulation promotes the disclosure of value-relevant accounting information. We further find evidence that analysts update their earnings forecasts and recommendations in accordance with textual changes in firm reports. This suggests that analysts contribute to market efficiency by conveying qualitative information from accounting statements to the public.