From Returns to Tweets and Back: An Investigation of the Stocks in the Dow Jones Industrial Average

ABSTRACT

A sizeable percentage of investors are using social media to obtain information about companies (Cogent Research [2008]). As a consequence, social media content about firms may have an impact on stock prices (Hachman [2011]). Various studies utilize social media content to forecast stock market-related factors such as returns, volatility, or trading volume. The objective of this article is to investigate whether a bidirectional intraday relationship between stock returns and volatility and tweets exists. The study analyzed 150,180 minute-by-minute stock price and tweet data for the 30 stocks in the Dow Jones Industrial Average over a random 13-day interval from June 2 to June 18, 2014 using a BEKK-MVGARCH methodology. Findings indicate that 87% of stock returns are influenced by lagged innovations of the tweets data, but there is little evidence to support that the direction is reciprocal, with only 7% of tweets being influenced by lagged innovations of the stock returns. Results further show that the lagged innovations from 40 percent of stock returns affect the current conditional volatility of the tweets, while 73 percent of tweets affect the current conditional volatility of stock returns. Moreover, there is strong evidence to suggest that the volatility originating from the returns to the tweets persists for 33 percent of stocks; the volatility originating from the tweets to the returns persists for 73 percent of stocks. Last, 53 percent of stocks exhibit both immediate and persistent impacts from returns to tweets, while 90 percent of stocks exhibit both immediate and persistent impacts from tweets to returns. These results may help traders achieve superior returns by buying and selling individual stocks or options. Also, asset and mutual fund managers may benefit by developing a social media strategy.

KEYWORDS:

• Twitter
• Dow Jones Industrial Average
• Volatility
• Stock returns

Notes

1. This is a random thirteen-day interval. Other thirteen-day intervals and intervals of different lengths provide virtually identical results.

2. For example, if prices are assumed to be distributed log normally, then log (1 + r_i) is normally distributed. Moreover, when returns are very small (common for trades with short holding durations such as these), the logarithm of the return relative is very close in value to raw returns.

3. The stock return outlier of Intel Corporation occurs at observation 3447 and requires a dummy in the variance space of the GARCH model.

4. Hull and White (1998) assert that a discrete mixture of normal distributions can be used to explain the observed patterns of significant kurtosis and a positive skewness of data.

5. The decaying processes of the data series are more easily seen on a graph but due to space constraint we decided to omit these graphs. However, the graphs are available upon request.