Predicting stock returns using large language models
Text is an underexploited data source for understanding asset markets. So far, academic research has studied only a tiny fraction of market-relevant text data and often focuses on a single specialised data source. As a result, text information is often represented in rudimentary ways. This report summarises a study that aims to improve stock-return prediction models by extracting contextualised representations of news text derived from large language models (LLMs).