How twitter informs the stock market

Research suggests Twitter trends can accurately forecast stock market changes.

On April 23, the US stock market suddenly lost 1.5% of its value after news of an attack on the White House quickly spread from the (hacked) Twitter account of the Associated Press news agency. Within a matter of minutes the fake news report was debunked and the market recovered from this “flash crash”. But whether the hack was perpetrated by the Syrian Electronic Army or by shrewd traders looking to make a quick profit, what emerges from this incident is how important social media has become for investors.

Gone are the days when information would spread from news agencies to television and newspapers, and from there to the public. The pyramid of access is now much flatter, with every (connected) soul on the planet able to access a wealth of information through social media. The key question to ask is whether all this access is actually useful. Is it possible to extract golden nuggets of financial wisdom from this deluge of news, opinions and investment tips?

Recent academic research seems to say yes. A number of authors have suggested that social media feeds can be used to successfully forecast equity returns. The idea is not to sift through the mass of views and tips looking for the “best expert”, but to aggregate all the available information into a comprehensive measurement that reflects how the market feels about a particular company.

Modern technology allows us to download, in real time, a vast amount of data in machine-readable format and to then analyse the semantic content of the text. Every tweet or post mentioning a stock can be read and categorised as “positive” or “negative” based on the words it contains. Investors look for stocks that are “trending”, with a high number of mentions, and with an overwhelming proportion of positive versus negative words.

Researchers at Purdue University have shown that it is possible to extract useful information from user-generated investment opinions on Seeking Alpha, one of the biggest investment-related social media websites in the United States. They have shown that the proportion of words with negative meaning accurately predicts the future return of the stocks in question. Similar results have been obtained using Twitter feeds and even aggregating in real time a variety of social and traditional media.

None of these studies proves that the profitability of these investment strategies can withstand significant transaction costs, but the results are still interesting. While there is overwhelming evidence that individual investors, on average, are unable to trade successfully, the aggregation of all of their opinions leads to a somewhat profitable trading strategy.

One possible explanation is that we are looking at the information-age equivalent of the traditional “wisdom of the crowds” phenomenon observed by British statistician Francis Galton. At a 1906 country fair in Plymouth, 800 people took part in a contest to estimate the weight of an ox. While the individual guesses where all over the place, the median estimate was remarkably accurate. Individual biases cancel out when there are a large number of independent guesses and the aggregate estimate becomes more accurate.

It may be appealing to liken the stock market to a big fat ox, but is this really what is going on here? A cornerstone of the “wisdom of the crowds” phenomenon is that individual estimates have to be “independent”. Only then can we expect individual errors to compensate for one another.

The Associated Press incident shows that social media does not promote independence – a single tweet by an influential source can be retweeted, liked and cross-posted thousands of times, snowballing into an avalanche of correlated comments.

An alternative explanation of this new phenomena is that the social media “mood” is forecasting not the correct value of a stock but the short-term trading intentions of the market. Algorithmic traders can use this information to “front-run” other investors by taking up positions in trending stocks.

The difference between these two explanations is not academic. In the first case social media would allow a quicker diffusion of valuable information and improve market efficiency. In the second case it would simply generate herding behaviour among investors and increase market volatility.

The jury is still out. But what is certain is that social media will become more and more relevant for sophisticated algorithmic traders and will increase their ability to front-run the very same market participants who kindly volunteered their trading intentions in their first tweet of the morning.