Enhancing Stock Price Forecasting with a Multi-Agent System and Time Series Analysis

Stock price forecasting remains a highly challenging task due to the dynamic and volatile nature of financial markets. Stock prices are influenced by a multitude of external factors such as global events, economic policies, and public sentiment. Traditional time-series models often fall short in capturing these complex dependencies as they primarily rely on historical price data. In this study, we propose a novel framework that integrates a multi-agent system powered by Large Language Models (LLMs) with traditional time-series forecasting models such as SARIMAX, LSTM, and a Time-MOE transformer-based architecture. The LLM-based agents are responsible for analyzing financial news and market data, rating content across key factors like financial stability, marketing strategies, and risk indicators. These insights are then combined with price history to provide a more context-aware prediction model. Experiments conducted on stock data from major technology companies show a significant reduction in Mean Absolute Error (MAE) compared to baseline models, demonstrating the effectiveness of our approach. The results emphasize the potential of combining agent-based sentiment evaluation with traditional models to improve forecasting accuracy in a scalable and cost-effective manner