A0503
Title: Investor sentiment and the cross section of stock returns: A natural language processing approach
Authors: Francesco Audrino - University of St Gallen (Switzerland)
Fabio Sigrist - ETH Zurich (Switzerland)
Jule Schuettler - University of St.Gallen (Switzerland) [presenting]
Abstract: The aim is to investigate how investor sentiment affects the cross-section of stock returns using a data-driven natural language processing (NLP) methodology. Daily sentiment is derived from various text data sources, including newspaper headlines, tweets from Stocktwits, and earnings call transcripts. A state-of-the-art NLP model is applied for sentiment classification, with labels generated based on one-day-ahead stock returns. The model's output can be interpreted as a one-day-ahead return forecast, which is utilized for conducting portfolio sorts. The contribution is twofold: first, sentiment is directly derived from text data, eliminating the reliance on proxies; second, labels are generated through a data-driven process rather than human annotation.
