FinRL Contest 2025 Task 1:Market-Aware In-Context Learning Framework for Proximal Policy Optimization in Stock Trading Using DeepSeek

This paper introduces a Market-Cap Stratified Subset (MCSS) strategy. It constructs a balanced, high-quality dataset from the Financial News and Stock Price Integration Dataset (FNSPID). Building on MCSS, we propose Market-Informed Sentiment for Trading (MIST). This is a two-phase framework that first generates sentiment scores through optimized prompt engineering with LLMs, then aligns market movements to deliver actionable trading signals. With a Proximal Policy Optimization (PPO)-based agent, MIST achieves better policy stability and learning efficiency than baseline methods.