Sentiment Analysis of Financial News Headlines with Market Comparison

Growth of the internet and the digital economy, along with technical advances in computer and data science have supported a wave of alternative data sources that can be used to measure and predict the financial markets. One of these non- traditional metrics is public opinion mining, commonly referred to as sentiment analysis. This study investigates the hypothesis that the sentiment of financial news headlines reflects and directs the performance of the U.S. stock market through proving a significant correlation between the polarity of the sentiment and the change in price of a security, thus working to disprove the controversial ‘efficient market hypothesis’. To evaluate the publics sentiment a vast dataset of ‘financial news’ headlines are required ranging over a broad period. Additionally, a natural language processing and machine learning classification model is built to predict the sentiment polarity of headlines. Finally, statistical analysis is conducted on the data to prove any significant correlation within the results. The study can demonstrate the hypothesis to an extent, showing that the sentiment of financial news headlines relating to the overall U.S. market, directly reflects the price of the U.S. market index. Despite this no correlation between the price of an individual stock and the sentiment of directly relating financial news headlines could be found. Additionally, there was no evidence to suggest that the daily sentiment of a security had any influence over its corresponding price change the subsequent day.