Reading Between the Reels: An AI ‐Driven Approach to Analysing Movie Review Sentiment and Market Returns

Abstract

This study examines the relationship between market sentiment derived from movie reviews and stock returns. Using GPT‐4o large language model (LLM), we construct daily sentiment measures from approximately 247,850 movie reviews. Empirical results indicate a robust negative relationship between movie sentiment and excess market returns, suggesting that well‐received theatrical films distract investor attention from fundamental market information, leading to reduced market participation and lower returns. The robustness of this finding is confirmed through a comprehensive set of validation tests, including out‐of‐sample forecasting, multiple testing corrections, and validation against human‐annotated sentiment scores. Time‐series regressions indicate that the negative relationship between movie sentiment and daily market excess returns persists for approximately 2 or 3 days, but does not hold over longer horizons, and no reversal in returns occurs thereafter. Heterogeneity analyses reveal that this effect is more pronounced during bear markets and the COVID‐19 pandemic period, highlighting investor behaviour driven by cognitive fatigue and emotional diversion during stressful market conditions. However, during financial crises, investor attention remains largely focused on critical macroeconomic events, reducing sensitivity to non‐financial sentiment signals. Further analyses confirm that elevated movie sentiment negatively affects trading volume and positively influences market volatility, consistent with the attention distraction hypothesis. Our findings remain robust across alternative deep learning architectures including BERT, LSTM, CNN, Bi‐LSTM and TIIF. Overall, the findings highlight the importance of investor attention to multimedia visual stimuli and psychological mechanisms in shaping financial market outcomes.