A Universal LSTM Stock Price Predictor Utilizing News Sentiment Analysis and Technical Indicators
Abstract
The stock market is influenced by a complex interplay of factors, including historical price trends, technical indicators, news sentiment, and macroeconomic conditions. Traditional stock prediction models typically focus on a single stock, limiting their ability to capture broader market relationships. Investors require a model that can accurately forecast price movements across multiple stocks to optimize trading decisions.
We propose a universal stock prediction model that leverages relationships across all S&P 500 stocks. Unlike traditional single-stock models, our approach utilizes a multi-stock LSTM architecture trained on a combination of historical stock prices, technical indicators, and news sentiment. The model is designed to capture hidden interdependencies between stocks, enabling it to predict how breaking news about one company may influence the prices of others.
Our model builds upon insights from baseline experiments, where we tested various input combinations to determine the most impactful features. By isolating different input categories in our baselines, we identified that stock prices, technical indicators, and news sentiment provided the highest predictive accuracy. The universal model integrates these factors and learns market-wide patterns to improve overall forecast reliability.
By incorporating market-wide learning, our approach aims to enhance short-term stock price predictions beyond traditional methods. We validate our model’s performance against multiple baselines, demonstrating its potential to help investors make more informed trading decisions.
