An Intelligent Stock Price Prediction System Using Large Language Models in Financial News Understanding

Abstract

The stock market is always fluctuating due to factors Numerous variables, including investor mood, economic statistics, and financial news, impact the ever-changing stock market. Natural Language Processing (NLP) has emerged as an effective approach for analyzing financial news headlines and identifying market sentiment to improve stock price forecasting. This study proposes an intelligent stock price prediction system using a Financial Generative Pre-Trained Transformer (FinGPT), which integrates financial news understanding with historical S&P 500 market data to predict stock closing prices more accurately. The data was cleaned, labelled, and normalized using Z-scores before model training, and the train-test split was 70:30. R³, RMSE, and MAPE evaluation measures were used to evaluate the performance of the proposed FinGPT model with ARIMA, LSTM, and GRU. As shown by the experimental findings, the suggested FinGPT outperformed the previous models in terms of prediction accuracy, with an R² of 99.9%, an RMSE of 39.9628, and a MAPE of 0.8267. The outcomes demonstrate that the suggested framework is an effective and trustworthy system for intelligent financial forecasting by combining the strength of transformer-based financial language understanding with historical market data. This integration significantly improves the accuracy of stock price prediction.