Essays On Investors Sentiment And Attention PDF Download

The first paper investigates the predictive power of investors' sentiment and attention for the stock returns' volatility. We introduce a novel and extensive dataset that combines information from social media platforms, news articles, search engine data, and information consumption. Applying a state-of-the-art sentiment classification technique, we construct measures of investors' sentiment and attention for 18 U.S. stocks and the financial market in general. We identify investors' attention, as measured by the number of Google searches on financial keywords (e.g. «financial market» and «stock market»), and the daily volume of company-specific short messages posted on the social media platform StockTwits to be the most relevant variables. The second paper investigates a potential driver of the predictive power documented in the first paper. We focus on news releases of 360 U.S. companies from the S&P 500 universe and analyze how investors' attention affects the speed at which new information is incorporated in stock prices. Our results show that higher investors' attention around news releases is related to higher contemporaneous volatility. Further, retail investor attention increases the post-announcement volatility, whereas institutional investor attention has a small but negative impact on volatility on days following news releases. The third paper extends the analysis of the first paper to the multivariate stock return volatility. Building on the theoretical and empirical evidence that links the price comovements with retail investors' behavior, we analyze the predictive power of retail investors' sentiment and attention for the realized correlation matrix of 35 Dow Jones stocks. We propose a new model of realized covariances that allows exogenous predictors to influence the correlation dynamics while ensuring the predicted matrices' positive definiteness. Using this model, we find retail investors' attention to have predictive power for return correlations, especially for longer forecasting horizons and during the COVID-19 pandemic. The last paper analyzes in more detail the time-series properties of the daily online investor sentiment measures used in the first two papers. We detect structural breaks in the sentiment series for most of the 360 U.S. companies considered in this paper. We illustrate the economic significance of this finding with a return prediction exercise.