Greece? Zzzz. EU bank says TWEETING can move the stock market
It must be true, Big Data says so
Worstall on Wednesday What is it that Twitter is better at than Google at doing? Over and above the obvious point that Twatter is better at broadcasting 140 character apercus to the world, while Google is better at telling you the answer to something?
That answer is, as the boys at the European Central Bank (PDF) have just told us all, predicting stock market rises. Perhaps this is why they've had such troubles in sorting Greece out, that they've been doing it with models instead of solving real-world problems.
Still, it is an interesting result. What they wanted to find out is, when everyone is sending twats to each other assuming that the stock market is going to go up, does it go up? And when those messages are gloomy, does it go down? Similarly, when the volume of searches on the idea that the market might go up rises, does the market subsequently rise?
Which is a bit of a surprise, really. What they did was search the Twat-feed for times when “bullish” and “bearish” increased in frequency and then compared that to the subsequent performance of the market:
We observe that a one standard deviation increase in Twitter bullishness on day t − 1 is followed by a 12.56 basis points increase in DJIA returns on the following day. This impact is statistically significant at the 99% confidence level. In addition, compared with the unconditional mean of daily Dow returns during the sample period, i.e. 3.46 basis points, a figure of 12.56 basis points is also economically significant.
That's actually a lot. Certainly it's enough to base a trading strategy upon, assuming you can gain access to the firehose of the information. Similar results were found when “bearish” rose in frequency, but obviously associated with falls in the market index. They checked this against the UK, Canada and US, all places with liquid markets and lots of Twitter users, and it worked. They also tested China, where there is no Twitter, and found no correlation with the markets: the null case.
There were similar but smaller effects when looking at the frequency of searches for “bull market” and “bear market” on Google. That information is only available on a weekly basis rather than Twitter's daily, so it's unknown as to whether the effect is lower itself or that it's being masked by the passage of time.
Their other finding is that the rise in the market from twatted optimism fell away again over the subsequent four days. So, weekly numbers might just be masking this.
Do note that we are talking about basis points here (0.01 of a per cent) so this isn't something a day trader is about to put to good use. But it quite probably is one of those things it's worth someone putting an algo onto and seeing how long it works before everyone does it and it gets arbitraged away – or, alternatively, the price rise comes immediately and not in the next day.