MarketSense: Do Small AI Models Understand Financial News?

Author

Sai Mahesh Sandeboina
Independent Researcher, USA
📧 saimaheshsandeboina931@gmail.com

---

Abstract

This study examines whether compact language models like FinBERT and DistilBERT can extract meaningful sentiment signals from financial news. Using headlines for nine major NASDAQ tickers (Apr–Oct 2025), we correlate daily sentiment with same-day stock returns and find that FinBERT—despite its small size—captures weak but significant relationships, suggesting that domain adaptation can outperform model size.

---

Repository Contents

• main.tex – full LaTeX source
• references.bib – BibTeX citations
• figures/ – all experiment plots
• sections/ – structured paper content
• artifacts_results/ – reproducible CSVs & correlation data
• MarketSense_Paper_SaiMaheshSandeboina.pdf – final compiled paper

---

Citation

If you reference this work:

@misc{sandeboina2025marketsense,
  title={MarketSense: Do Small AI Models Understand Financial News?},
  author={Sai Mahesh Sandeboina},
  year={2025},
  note={arXiv preprint (to appear)}
}
---

### Keywords
`Finance AI` • `LLM Evaluation` • `FinBERT` • `DistilBERT` • `BloombergGPT` • `NVIDIA Research` • `Market Sentiment`