Using Textual Information in Econometrics: Quantifying Newspaper Sentiment

This paper uses artificial intelligence text analysis methods to construct indices of economic sentiment from a database of 47,000 articles from the Financial Times. The indices have high explanatory power for predicting Federal Open Market Committee interest rate decisions; the effect is both statistically and economically significant. This is partially explained by the incremental predictive power for economic growth the measures exhibit even when accounting for FOMC Greenbook forecasts. However, the FOMC is found to respond strongly even to uninformative components of newspaper sentiment. The result is therefore similar to Romer and Romer (2008), who find that the FOMC reacts to uninformative private forecasts.