Reproduce “TRACE: Temporal Rule-Anchored Chain-of-Evidence on Knowledge Graphs for Interpretable Stock Movement Prediction” 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
Qianggang Ding, Haochen Shi, Luis Castejón Lozano, Miguel Conner, Juan Abia, Luis Gallego-Ledesma, Joshua Fellowes, Gerard Conangla Planes and 2 more
The TRACE framework was developed to predict daily stock movements using a temporal financial knowledge graph, integrating rule-guided exploration and LLM-validated reasoning to provide interpretable UP/DOWN verdicts. It achieved 55.1% accuracy and a 41.7% total return in backtesting, outperforming strong baselines while generating human-readable explanatory paths grounded in textual evidence.
Key insights
- Combining symbolic relational rules with LLM-guided graph exploration in a temporal knowledge graph can achieve both improved predictive performance and true interpretability in financial forecasting.
- Explicitly grounding reasoning paths in real-world news articles and structured corporate relations, combined with temporal validity constraints, is crucial for generating auditable and economically meaningful explanations.
- Multi-hop reasoning, rule mining, and temporal constraints are essential components that significantly contribute to the accuracy and robustness of stock movement predictions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.