Correlating tweets with the daily return reversal of the US stock market

Abstract

In the stock market, return reversal happens generally due to the fact that investors sell overbought stocks or buy oversold stocks, reversing the stocks' prices. In this paper, we aim at exploring correlations between tweets (posts on Twitter) and the daily return reversal of the US stock market. First, we employ dynamical Bayesian factor graph to depict dynamic interrelationships among a comprehensive set of tweet factors and economic factors related to the return reversal. Then, we build two models based on support vector machine (SVM) to predict the return reversals in a sliding time window. Empirical experiments demonstrate that tweet sentiment factors relate to the return reversal intimately. In specific, low level positive sentiment or high level negative sentiment in tweets leads to the return reversal with greater probabilities, and these sentiment factors have the ability of significantly improving the prediction of the return reversals.