Reproduce “Ploutos: Towards interpretable stock movement prediction with financial large language model” 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
Hanshuang Tong, Jun Li, Ning Wu, Ming Gong, Dongmei Zhang, Qi Zhang
Microsoft researchers developed Ploutos, a framework for interpretable stock movement prediction that combines a financial large language model with specialized expert modules. The system processes multimodal data to generate accurate stock predictions accompanied by detailed, ranked rationales, achieving higher accuracy and faithfulness compared to existing methods on benchmark datasets.
Key insights
- The multi-expert architecture, combining technical, sentiment, and human analysis, effectively integrates diverse market perspectives for improved prediction.
- Rearview-Mirror Prompting generates high-quality labeled training data by prompting LLMs to explain past stock movements based on prior information, enhancing model interpretability.
- Dynamic Token Weighting emphasizes critical information in rationales during fine-tuning, ensuring key insights receive appropriate focus and improving the quality of explanations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.