Structured Event Representation and Stock Return Predictability
Researchers show that large language models can extract structured event representations from financial news, boosting stock return predictions. Their method breaks news into subject-action-object triplets, making results more transparent than traditional sentiment analysis. The model, using attention mechanisms and knowledge graphs, achieved a 10.93 percent annualized return and a Sharpe ratio of 0.78. It reveals which events drive markets but depends on high-quality news and focuses on large-cap US stocks, leaving its broader applicability uncertain.
What it examines
This paper uses large language models to extract structured event features from financial news, aiming to improve stock return prediction. By turning news into human-readable event triplets and applying attention-based deep learning, the study seeks to make predictions more accurate and interpretable.
What it concludes
The event-driven model outperforms traditional text-based methods and offers clear insights into what drives stock returns. This approach can be applied to other assets and financial topics, making predictions more trustworthy and understandable. Future research may extend this method to new data sources and financial decisions.
Evidence objects
Researchers reveal that large language models extracting structured event triplets from financial news dramatically boost stock return predictions, offering far greater transparency than traditional sentiment or word-embedding approaches.
key_findings bullet 1 · key_findings · validation V0
The novel SER model, powered by attention mechanisms and knowledge graphs, achieved an impressive annualized return of 10.93% and a Sharpe ratio of 0.78, outperforming existing text-based financial prediction benchmarks.
key_findings bullet 2 · key_findings · validation V0
Despite its interpretability and predictive strength, the model depends on high-quality news and large-cap U.S. stocks, raising questions about its effectiveness for less-covered firms, other markets, or alternative news sources.
key_findings bullet 3 · key_findings · validation V0
This paper introduces a novel method using large language models to extract structured event representations (SERs) from financial news for stock return prediction. By employing subject-action-object triplets and attention mechanisms, it advances interpretability in financial NLP. Empirical results show significant outperformance and transparent attribution, making it compelling and original.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
Abstract: We find that event features extracted by large language models (LLMs) are effective for text-based stock return prediction. Using a pre-trained LLM to extract event features from news articles, we propose a novel deep learning model based on structured event representation (SER) and attention mechanisms to predict stock returns in the cross-section. Our SER-based model provides superior performanc… ▽ More We find that event features extracted by large language models (LLMs) are effective for text-based stock return prediction. Using a pre-trained LLM to extract event features from news articles, we propose a novel deep learning model based on structured event representation (SER) and attention mechanisms to predict stock returns in the cross-section. Our SER-based model provides superior performance compared with other existing text-driven models to forecast stock returns out of sample and offers highly interpretable feature structures to examine the mechanisms underlying the stock return predictability. We further provide various implications based on SER and highlight the crucial benefit of structured model inputs in stock return predictability. △ Less
Source row: 1873 · abstract type: unknown