API Augmented Reinforcement Learning Framework Utilizing LLMs for Enhanced News-Based Stock Portfolio Strategies

Abstract

This paper introduces an API Augmented Reinforcement Learning (RL) framework that utilizes Large Language Models (LLMs) to enhance stock portfolio strategies by leveraging real-time news data. Traditional financial strategies primarily focus on historical and technical indicators; however, our approach integrates sentiment analysis and summarization of news articles to influence reinforcement learning agents’ decision- making. News-based features, combined with market indicators, form a comprehensive state representation for training RL algorithms such as Proximal Policy Optimization (PPO). This hybrid system optimizes stock allocation dynamically, offering enhanced portfolio performance when evaluated against benchmark strategies.