'Big data' predicts stock movements, boffins claim
Google Trends is better than a dart board at stock-watching
Researchers from the Warwick University's business school reckon they can predict the next crash, by watching Google searches and Wikipedia.
The supposed power of Google as a predictive tool has been popular in academia ever since The Chocolate Factory unveiled Google Flu Trends back in 2008.
Its usefulness for real-world prediction was called into question earlier this year, when researchers concluded that it's almost always wrong.
That hasn't stopped the University of Warwick researchers from trying the same approach to financial prediction. In this paper at PNAS, the researchers explain that rather than picking the search vocabulary they're watching ahead of time (as previous work has), they've tried to identify the emergence of phrases on Wikipedia which have, in the past, been associated with subsequent crashes.
“First, we take a large online corpus, Wikipedia, and use a well-known technique from computational linguistics to identify lists of words constituting semantic topics within this corpus,” the study states.
The researchers then recruited Mechanical Turk users to give names to these topics, and examined Google searches from 2004 to 2012 to see if “the search volume for each of these terms contains precursors of large stock market moves”.
They believe that “for complex events such as financial market movements valuable information may be contained in search engine data for keywords with less-obvious semantic connections”.
In other words, what the researchers believe they've found is a connection between what people looking at the stock market are interested in before they've made a buy-sell decision – and that the technique could be useful in other applications. “We suggest that extensions of these analyses could offer insight into large-scale information flow before a range of real-world events”, they write.
It should be noted that rather than prying into actual queries, the researchers focussed on Google Trends' aggregations of search terms.
And it should also be noted that Google-as-hive mind only thrashed “random strategies” for predicting stock movements: there was only marginal support in the data that Google-watching might outperform a simpler buy-and-hold strategy. ®
Sponsored: Benefits from the lessons learned in HPC