LLM-Based Hierarchical RL for Pair Trading
A new arXiv preprint (2605.01954) introduces Moira, a language-driven hierarchical reinforcement learning framework for pair trading. The approach uses large language models (LLMs) to parameterize both high-level and low-level policies, optimized through prompt updates. It addresses credit assignment in hierarchical decision-making by leveraging trajectory- and episode-level textual feedback.
Key facts
- arXiv paper 2605.01954 introduces Moira
- Moira is a language-driven hierarchical reinforcement learning framework
- Applied to pair trading
- Uses LLMs for both high-level and low-level policies
- Policies optimized through prompt updates
- Uses trajectory- and episode-level textual feedback
- Addresses credit assignment in hierarchical decision-making
- Published on arXiv
Entities
Institutions
- arXiv