Investigating Stock Prediction Using LSTM Networks and Sentiment Analysis of Tweets Under High Uncertainty: A Case Study of North American and European Banks

Abstract

The investigation focused on the prediction of stock prices amidst significant macroeconomic and geopolitical volatility, particularly targeting North-American and European banks in 2022 — a year marked by intense economic shocks, including inflation, geopolitical tensions, and supply chain disruptions. A multidimensional approach was employed, integrating advanced Artificial Intelligence (AI) techniques such as Recurrent Neural Networks (RNNs) and sentiment analysis, utilizing a comprehensive dataset that includes traditional financial metrics and sentiment-driven data from social media, specifically Twitter (recently renamed X). By employing LSTM and FinBERT models, the study revolved around several key analyses: assessing the impact of different market conditions across the US and EU; exploring the potential benefits of data aggregation from multiple banks within these markets; examining the influence of varying historical data spans on model performance; and integrating sentiment analysis to capture the nuanced influence of public sentiment on stock movements. The findings indicate that market-specific dynamics significantly affect the predictive models, with higher inter-bank correlation observed in the US compared to a more fragmented European market. Additionally, models incorporating recent data streams and public sentiments tend to outperform those relying on traditional, longer historical data.

Keywords

• Artificial intelligence
• natural language processing
• sentiment analysis
• stock price forecasting
• time series