Bitcoin cryptocurrency is on fire. The value of a single bitcoin has soared from approximately $500 in 2014 to $6,000 (at times, as much as nearly $18,000).
But can that value can be manipulated by public sentiment, as expressed through social media channels?
"Common sense says that perhaps it can," says Stevens School of Business professor Feng Mai, who led a four-university team investigating the question.
Now Mai's group has confirmed that hypothesis — and discovered a surprise: how certain types of commenters carry up to ten times more weight as others.
Natural-language processing to sort social media data, statistical tests to analyze it
Mai, working with the University of Cincinnati, Dickinson College and Ivey Business School, collected and analyzed two years' worth of forum posts on the world's most popular public bitcoin forum, Bitcointalk.
The team collected the comment data using a Python script, classifying it into positive, negative and other sentiment categories using natural-language processing techniques. They also collected two months' worth of Twitter data, including more than 3.4 million tweets about Bitcoin.
Using a powerful statistical method known as vector error correction, or VECM, Mai's team then compared changes in Bitcoin's price with the chatter around the cryptocurrency. To account for broader economic effects, the team also factored in daily rises and falls in indicators such as the S&P 500 stock index, gold prices and volatility indexes.
The group chose VECM because it's useful for exploring large-scale questions such as the causes of GDP and currency valuation changes, which often possess dynamic relationships among their variables.
"It's not a one-way relationship," notes Mai. "Any changes in Bitcoin's price are obviously going to affect the sentiment around it, so we needed to factor in those influences as well."
The team's conclusion, recently published in the Journal of Management Information Systems, is that periods of increasingly positive social media commentary do in fact influence the rising price of Bitcoin significantly.
"Many of us probably intuitively believe this, but this was the first robust statistical finding to verify that social media and Bitcoin prices are actually linked," Mai says.
The power of the silent majority
The Stevens-led team didn't stop there. Next it took the research an additional step by dividing Bitcoin tweeters and posters into two groups: those who were posting very frequently (a small group that produced 60% of the total posts sampled) and those who were not.
"We did this because we are interested in what kinds of commenters affect prices most," Mai explains. "Vocal users of social media may sometimes have a certain agenda, in this case hyping or boosting the price of Bitcoin because they themselves have invested in it. So if most of the social messages around Bitcoin are generated by people who are biased, the sentiments on social media may not actually accurately reflect the currency's actual value.
"We wanted to know who is affecting the price: a vocal minority, who may be biased, or the quieter majority, who do not seem to have a reason to be untruthful, or both."
Indeed, this second analysis appears to demonstrate that Bitcoin investors recognize these potential conflicts of interest and discount them.
Very active users' comments and tweets, Mai's team found, did not move Bitcoin's price much at all. However, the "silent majority" — infrequent Twitter and Bitcointalk users who took the time to comment on the cryptocurrency's prospects — moved prices more, as much as ten times more, when they posted positive comments.
"This was a big finding, and it does seem to prove that people are trusting the silent majority much more, perhaps because they do not seem to have an agenda," says Mai.
"It's also true that, by following the infrequent posters' comments, you get a much more accurate prediction of Bitcoin's price over time, and this is useful for investors and potential investors to know."
Mai next hopes to investigate blockchain technology and its relationship to social media using similar techniques, he says.