LLM judges evaluate agentic stock predictors, improving accuracy via reinforcement learning

Researchers have developed a novel framework for evaluating agentic stock prediction systems by utilizing large language models as judges. This system breaks down performance into six specific dimensions, including regime detection and risk calibration, offering a more nuanced assessment than traditional aggregate metrics. The LLM judges, including GPT 5.4, Claude 4.6 Opus, and Gemini 3.1 Pro, demonstrated high agreement and correlated well with realized trading performance. This behavioral evaluation was then integrated into a reinforcement learning feedback loop, leading to significant improvements in prediction accuracy and trading strategy. AI

IMPACT Introduces a new method for evaluating and improving AI agents in complex decision-making tasks like financial prediction.

RANK_REASON Academic paper detailing a new evaluation framework for AI systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Claude 4.6 Opus
• Gemini 3.1 Pro
• GPT 5.4
• LLM judges
• Soft Actor-Critic

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →