Hiding in Plain Sight : Sentiment Analysis and the Efficient Market Hypothesis

Abstract

The stock market is a notoriously complex and unpredictable system, and because of this has always been an alluring subject for academic research seeking to make the unpredictable more predictable. This major research project is no different as it aims to quantify the predictive value of financial sentiment, determine which sentiments are most meaningful, when they are most meaningful, and if meaningful sentiment varies depending on type of stock. To pursue these goals, the project finds its theoretical footing in Eugene Fama’s Efficient Market Hypothesis and Daniel Kahneman’s Prospect Theory. However, the methodological component of this project enters into emerging territory as it employs sentiment analysis and machine learning, which have only recently been made possible by advances in technology and communications practices. Specifically, through the use of the Loughran-McDonald dictionary for financial sentiment, corporate press releases were analyzed and tested using a Random Forest machine learning model. The results from this project show that financial senitiment found in press releases does provide a slight predictive edge, however the sentiments responsible for that edge vary based on type of stock, type of fluctuation being predicted, and timeframe.