Investigating whether tweets related to the Federal Open Market Committee (FOMC) around the time of FOMC meetings contain important information about equity asset prices, the authors find that the market sentiment inferred from these tweets can be used to predict aggregate stock market returns, even after controlling for common market factors. Using simulated data, they also demonstrate that certain trading strategies that use this tweet content can outperform the market on multiple performance metrics.
Using a database of tweets that contain language related to the Federal Open Market Committee (FOMC), the authors document that the tweet content can be used to predict aggregate equity market results, even after controlling for common market factors. This finding suggests that the tweets contain valuable asset pricing information not conveyed by common market factors.
The authors assess the investment impact of using the contents of FOMC-related tweets by constructing simulated trading strategies that invest more or less in the equity markets based on the investor sentiment inferred from the tweets. Results indicate that portfolios that use the tweet information can significantly outperform passive buy-and-hold strategies.
How Is This Research Useful to Practitioners?
Practitioners’ interest in the informativeness of financial market–related content posted on such social media platforms as Twitter has grown in recent years. The results of this study follow those of previous researchers who suggest that such content can be invaluable and may not be reflected in common market risk factors.
As demonstrated by the authors through simulations, incorporating this tweet content into their investment decision-making processes may enhance practitioners’ portfolio returns. It is important to note that during the study period of 2009–2014, the equity market was rising and interest rates were historically low. The authors point out that the results may be different when equity markets are declining and interest rates have reverted to historical levels.
How Did the Authors Conduct This Research?
Using the Topsy API, the authors gather all English language tweets over the 2009–14 period that include the terms “FOMC,” “Federal Reserve,” or either “Bernanke” or “Yellen” (depending on who was the Fed chair at the time of the tweet). Using the Python Pattern algorithm, they assign individual daily tweets a polarity score between –1 and +1 to classify the sentiment of each tweet (where +1 is purely positive and –1 is purely negative). For each day, an overall measure of sentiment is estimated by taking a weighted average of individual daily tweet polarity scores (weighted by each user’s number of followers).
In a series of regressions, the daily excess returns of the CRSP Value-Weighted Index are regressed on the lagged (day t – 1) weighted average daily polarity score; an FOMC meeting date dummy variable; an interaction FOMC dummy/polarity score variable; and the size, value, and momentum market factors. Regression results indicate that lagged tweet polarity can be used to predict market excess returns, and this effect intensifies on days when the FOMC meets. Even after controlling for the three market factors, lagged tweet polarity remains statistically significant, suggesting that tweet content contains valuable asset pricing information not conveyed by common market factors.
The authors then assess the practical value of using the tweet polarity information in making investment decisions. They construct a number of simulated portfolios to show that levered portfolios that use this tweet information outperform levered portfolios that use only market returns and the FOMC event day indicator as predictors.
The number of social media users has surged in recent years, making the impact of social media content on asset prices an interesting question worthy of further research. Of course, because of the risks of data mining and the subjectivity of variables and factor assignment, any test of this relationship must be carefully evaluated for its statistical integrity. But the results of this study suggest that some social media content can contain information that is not already reflected in common market factors. A natural extension of this research would be to examine the impact of sentiment implied by social media content on the returns of individual stocks and other financial assets.