Using Sentiment Analysis to Understand Monetary Policy Uncertainty

This paper proposes a novel method to identify monetary policy surprises and analyze the impact of monetary policy uncertainty by applying aspect-based sentiment analysis in natural language processing on documents from Federal Open Market Committee (FOMC). With techniques in machine learning, I associate each document with a sentiment score and categorize monetary policy based on how hawkish/dovish it is. In particular, by assessing the difference in sentiment scores of FOMC statements and Minutes that are released weeks apart, I construct the monetary policy uncertainty index, and find that the reaction of long-term yields is higher/lower when uncertainty is low/high.