Abstract
In behavioural economics, Big Social Data plays a crucial role in predicting stock market trends. This study aims to compare the effectiveness of the VAR model and the LSTM neural network in forecasting the relationship between social media and stock markets. Two hypotheses guide this work: first, verifying a statistically significant link between Twitter (X) activity and stock market metrics, and second, assessing the relative accuracy of the methods. Sentiment analysis, using both lexicon-based (VADER, NRC) and supervised learning (Naïve Bayes), was applied to construct affective indicators from textual data. Findings suggest that Twitter activity holds predictive value for trading volume and closing prices, though effects vary across timeframes and methods. Both VAR and LSTM yield stable insights over shorter periods. This analysis, focused on Apple and Amazon during 2016–2017, contributes to methodological advancements in exploring social media's impact on financial markets.