Studying about 1,200 economy-related tweets of President Trump, we establish
the “President reacts to news” channel of stock returns. Using high-frequency
identification of market movements and machine learning to classify the topics
and textual sentiment of tweets, we address the observed heterogeneity in the
aggregate stock market response to these messages. After controlling for market
trends preceding tweets, we find that 80% of tweets are reactive and predictable
rather than novel and informative. The exceptions are trade war tweets, where the
President has direct policy authority, and his tweets can reveal investable private
information or information about his policy function.
