Scholarly Comments on Academic Economics

Shy of the Character Limit: “Twitter Mood Predicts the Stock Market” Revisited

by and

Read this article

Access statistics
1,079 article downloads
905 complete issue downloads
Total: 1,984

Abstract

In the 2011 article “Twitter Mood Predicts the Stock Market” by Johan Bollen, Huina Mao, and Xiaojun Zeng, published in Journal of Computational Science, the authors estimated a proprietary measure of Twitter ‘calm’-ness and found that this measure Granger-caused increases in the Dow Jones Industrial Average from February 28, 2008 to November 3, 2008. The paper and related work has attracted attention by journalists, finance practitioners, and academics. In just six years the paper accumulated 2,514 Google Scholar citations and 678 Web of Science citations (as of April 24, 2017). We subject the paper to thoroughgoing scrutiny, including an attempt to replicate the findings in-sample. Using the authors’ subsampling technique, we replicate their sample window (less the since-deleted and hence unavailable tweets) and extend it backwards to include 2007. Constructing multiple measures of Twitter mood using word-count methods and standard sentiment analysis tools, we are unable to reproduce the p-value pattern that they found. We find evidence of a statistically significant Twitter mood effect in their subsample, but not in the backwards extended sample, a result consistent with data snooping. We find no evidence that our measures of Twitter mood aid in predicting the stock market out of sample. Congruously, the hedge fund set up in late 2010 to implement the Twitter mood strategy, Derwent Capital Markets, failed and closed in early 2012.