Investor sentiment and market returns: A multi-horizon analysis

Abstract

This study explores the relationship between investor sentiment and market return in the stock market, spanning both long-term and short-term horizons. Using a decade-long dataset (2013–2023) from Facebook, comprising around 773,000 curated posts from an initial 900,000, the research employs the Vector Error Correction Model (VECM) to illuminate long-run dynamics, revealing an equilibrium-restoring mechanism post-shocks between investors’ sentiment and Vietnamese stock market index (VNIndex). Short-term insights emerge from logistic and quantile regression analyses, categorizing market returns based on sentiment and elucidating relationships across market return distribution quantiles. The study also applies advanced machine learning algorithms—such as Decision Tree Regression (DTR), Support Vector Machine (SVM), Neural Networks (NN), Gradient Boosting Machine (GBM), Random Forest (RF), and Deep Neural Networks (DNN)—to demonstrate the predictive power of sentiment indices in forecasting abnormal returns on the VNIndex. The results emphasize the paramount influence of investors’ sentiment in terms of its predictive power compared to traditional autoregressive models of past trading data. Distinct patterns arise when comparing the low and high quantiles of returns distribution, with sentiment indicators being more influential at the lower quantiles. In summary, the research underscores the significant role of investor sentiment in the Vietnamese stock market dynamics and highlights the confluence of sentiment analysis and modern machine learning as a promising frontier in financial research.