The Power of Sentiment and Attention: Twitter and Google Trends as predictors of stock returns

Abstract

This thesis examines the impact of Twitter sentiment and Google Trends-derived investor attention on stock returns of Apple, Amazon, Microsoft, and Tesla. Spanning from January 1, 2018, to February 1, 2023, we extracted 0.55 million tweets, constructed a daily Google Trends Search Volume Index (SVI), and gathered adjusted close prices. Sentiment classification is done through a dual approach, integrating lexicon-based classification with machine-learning. By setting up multiple VAR models on first differences, we compare them to a random walk with drift and an AR model based solely on stock lags. We consistently outperform the random walk, challenging the efficient market hypothesis. Compared to the AR model, the results were mixed across the stocks, suggesting that investor behaviour may contribute to market inefficiencies for certain stocks. Our results suggest limited predicting power for Sentiment and SVI. SVI shows slight superiority over sentiment in predictive power, although its impact remains limited, with stock price fluctuations predominantly tied to their historical performance.