---
title: From Headlines to Holdings: Deep Learning for Smarter Portfolio Decisions
url: https://www.ml-quant.com/papers/arxiv/2509.24144/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-09-26
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2509.24144
source_url: http://arxiv.org/abs/2509.24144v1
featured: 2025-10-03
citations: 0
topic: Portfolio & Allocation
---


# From Headlines to Holdings: Deep Learning for Smarter Portfolio Decisions

A new deep learning framework for portfolio optimization, which combines LSTM networks, Graph Attention Networks, and financial news sentiment analysis, outperforms traditional benchmarks by directly learning portfolio weights, resulting in higher cumulative returns and Sharpe ratios.

- Source: http://arxiv.org/abs/2509.24144v1
- Identifier: arXiv:2509.24144
- Released: 2025-09-29
- First featured: Quant Letter No. 114 (2025-10-03): https://www.ml-quant.com/issues/2025-10-03/
- Citations (Semantic Scholar): 0
- Published in: not yet
- Topic: Portfolio & Allocation

## Related

- [Stock Movement Prediction and Portfolio Optimization](https://www.ml-quant.com/papers/ssrn/5234375/): The study introduces a deep learning framework, MR2TNet, to predict NIFTY50 index stock price movements by considering multiple information sources.
- [Graph Attention Networks for Portfolio Optimization](https://www.ml-quant.com/papers/ssrn/4757308/): The paper finds that Graph Attention Networks yield better investment outcomes compared to traditional complex system-based optimization models.
- [Optimizing Portfolio Allocation](https://www.ml-quant.com/papers/ssrn/4534131/): The BlackLitterman model, BLEnd2End, uses deep learning to optimize portfolio allocation, outperforming mean-variance benchmarks and other traditional strategies.
- [Signed network models for portfolio optimization](https://www.ml-quant.com/papers/arxiv/2510.05377/): The study shows that using negative edges in weighted signed network representations of financial markets can help reduce portfolio risk, performing on par with traditional models.
- [Bayesian Portfolio Optimization by Predictive Synthesis](https://www.ml-quant.com/papers/arxiv/2510.07180/): The article discusses a portfolio optimization technique using Bayesian predictive synthesis, which merges various asset return prediction models to handle financial market uncertainty.
- [Diffusion-Augmented Reinforcement Learning for Robust Portfolio Optimization under Stress Scenarios](https://www.ml-quant.com/papers/arxiv/2510.07099/): The research introduces a framework called DARL that combines DDPMs with DRL for portfolio management, improving its ability to withstand crises.