Reproduce “FinMarBa: A Market-Informed Dataset for Financial Sentiment Classification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Baptiste Lefort, Eric Benhamou, Beatrice Guez, Jean-Jacques Ohana, Ethan Setrouk, Alban Etienne
FinMarBa introduces a financial sentiment dataset constructed using a market-driven annotation framework, overcoming limitations of traditional human-labeled datasets by directly linking sentiment to observable asset price movements. The dataset, comprising 61,252 headlines, enables a BERT model fine-tuned on FinMarBa to achieve a Sharpe Ratio of 0.30 in predicting S&P500 movements, outperforming FinBERT's -0.13.
Key insights
- Directly linking sentiment labels to actual market reactions eliminates human biases and inconsistencies, yielding more objective and reliable financial signals.
- Automating the annotation process with LLMs like GPT-4 enables the creation of large-scale, high-quality datasets that are essential for training and evaluating modern language models in finance.
- Using trusted, globally relevant sources like Bloomberg Market Wraps, combined with market-driven labeling, produces a dataset that better reflects real-world financial dynamics and macro signals.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.