No such thing as negative press? – How negative media attention on euro area countries affects the yields of their government bonds

Abstract

Natural language processing and more specifically sentiment analysis has gained popularity as a viable method in finance research. Much of this research has focused on the stock market. In this thesis, I analyze the connection between news sentiment about EMU-countries and the yields of those countries’ 10-year bonds. To assess news sentiment I create a dataset of news headlines by scraping data from Twitter, and measure the adjusted proportion of negative words in them. I find that negative news is associated with statistically insignificant increases in bond yields right after the news come out. Based on my analysis I highlight the need for controlling for the novelty of news and propose a rudimentary metric to proxy novelty. I further discuss the inherent difficulties of news as a data source, especially in combination with time series data and outcomes that change slowly across time.