Integrating Sentiment and Technical Analysis with Machine Learning for Improved Stock Market Predictions

Abstract

This thesis advances stock forecasting by integrating sentiment analysis from Twitter as social media platform with traditional technical indicators, employing machine learning (ML) techniques. The research identifies gaps in existing literature, particularly in the use of appropriate validation methods and the balance of statistical metrics with financial benchmarks. It proposes a comprehensive methodology that incorporates Time Series Cross- Validation and hyperparameter tuning to enhance the adaptability and economic robustness of forecasting models. The empirical analysis unfolds in three chapters: 1. Technical Analysis within LSTM models to predict movements of the SPY ETF, validated through Time Series Cross-Validation to ensure robustness, focusing on both accuracy and financial performance. 2. Integration of Sentiment Analysis to assess its impact on model responsiveness and financial outcomes, demonstrating improved predictive accuracy. 3. Application to a Diverse Stock Portfolio, where models are applied to 10 different stocks across various sectors, confirming the models’ effectiveness and practical utility in real-world trading strategies. Key findings suggest that incorporating sentiment analysis significantly enhances the predictive precision of models, particularly in volatile market conditions. This synergy between technical indicators and sentiment data not only boosts accuracy but also enriches the models’ economic performance, offering valuable insights for traders and academic researchers exploring complex financial markets.