Economic Prediction with the FOMC Minutes: An Application of Text Mining

Abstract We conduct a sentiment analysis of the FOMC (Federal Open Market Committee) minutes based on the text mining results and examine the predictive ability of the resulting sentiment indicators. An adaptive Bayesian approach is employed to build the sentiment indicator for each of the Fed's mandates. We also improve existing mining techniques by identifying economics-related compound words and terminology in the minutes. Our empirical study shows that the mandate-specific indicators exhibit distinct patterns which help illustrate the FOMC's policy emphasis in different periods. It is also shown that these indicators are useful in predicting economic variables and generating superior out-of-sample forecasts. These results support the existing findings that the Fed possesses valuable information about the U.S. economy.