Predicting the Fed Sentiment in a Data-Rich Environment

Abstract

This paper examines the predictability of the Federal Reserve (Fed) sentiment conveyed by the words in the Federal Open Market Committee (FOMC) statements. First, we construct a Fed sentiment index based on textual analysis. Second, we predict the Fed sentiment index by using a large set of macro-finance variables (a data-rich environment). For the prediction, we employ several methods; OLS regressions, factor models, and penalized regressions. We find that most of the models outperform the benchmark, AR (1), in our in-sample analysis, suggesting that the use of a large dataset can improve forecasting performance. However, a simple OLS model with economic policy uncertainty and industrial production is the only model that beats the benchmark in our out-of-sample setting. Moreover, we assess the usefulness of the predicted Fed sentiment by forecasting two financial variables. The results suggest that the predicted Fed sentiment indices provide more information than its lag.