Market responses to sentiment shocks: A machine learning approach to major sporting events

This study investigates how major U.S. sporting events influence stock market performance by integrating traditional event study methodology with a machine learning-based counterfactual framework. The analysis relies on a dataset encompassing over 200 event-years across nine of the most prominent U.S. sports competitions between 2000 and 2024. Beyond standard market variables, the paper incorporates rich event level sentiment proxies including national television viewership, Google Trends search intensity, competitive intensity, and advertising costs. The empirical results demonstrate a significant divergence between models, as the machine learning specification identifies distinct pre-event anticipatory dynamics consistent with the emotional anticipation hypothesis. These are followed by stronger and more persistent abnormal return patterns extending 30 trading days post-event, whereas the classical Market Model captures only transient post-event reactions. Results align with behavioral finance theories that view emotionally salient public spectacles as exogenous sentiment shocks generating predictable mispricing and a multi-phase sentiment cycle of attention buildup, peak engagement, and gradual correction. By demonstrating the superior sensitivity of non-linear ensemble methods in estimating counterfactual expected returns, the study advances event study methodology and offers actionable insights for market surveillance and investor education during periods of heightened public attention.