Reproduce “Janus-Q: End-to-End Event-Driven Trading via Hierarchical-Gated Reward Modeling” 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
Xiang Li, Zikai Wei, Yiyan Qi, Wanyun Zhou, Xiang Liu, Penglei Sun, Jian Guo, Yongqi Zhang and 1 more
Janus-Q, an end-to-end event-driven trading framework leveraging large language models, integrates financial news events as primary decision units through a hierarchical-gated reward model. The framework achieved a Sharpe Ratio of 1.3088, outperforming 16 baselines, and reduced prediction Mean Absolute Error by 18.3% compared to the best time-aware model.
Key insights
- Directly optimizing the event-to-decision mapping within an LLM framework, rather than treating textual information as merely auxiliary, is crucial for achieving robust and profitable financial trading.
- A structured, hierarchical reward mechanism that explicitly models trade-offs among multiple financial objectives (e.g., directional accuracy, event-type consistency, PnL magnitude) significantly improves learning stability and decision-oriented performance.
- A comprehensive, expert-annotated event-centric dataset that unifies event types, market impacts (CAR), and semantic interpretations is essential for benchmarking and developing advanced event-driven financial models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.