SENTIMENT ANALYSIS OF NEWS ARTICLES: IMPLICATIONS FOR STOCK RETURNS PREDICTION AND MARKET TURBULENCE FORECAST

Abstract

This paper models sentiment on both an individual stock level and a general market level using sentiment analysis methods such as Latent Dirichlet Allocation and Gibbs sampling approximation. News article summaries on specific US stocks are used as inputs to generate sentiment indices using the sentiment analysis model built in this paper. These indices are then incorporated into the Fama French Three-Factor Model for stock returns, and are shown to have statistically significant correlations with excess returns on stocks. Market sentiment indices are then constructed from aggregating individual stock sentiment indices. By examining market sentiment indices and the Chicago Board Options Exchange Market Volatility Index (VIX), a statistically significant correlation is shown, which poses the possibility for sentiment modeling to be utilized by regulatory organizations in the forecasting of financial turbulence.