Stock exchange fluctuations are directly linked to investor mood and sentiment vis-à-vis the economy and the markets. What if such fluctuations could actually be predicted?

Google Search Information to Help Predict Stock Market Movements?

A team of researchers from the Warwick Business School in the UK and Boston University in the US have found a means of identifying which Google search topics herald dips in the stock markets. This research follows on from similar work already published, based on the audience for Wikipedia pages. Here once again the researchers have focused on the exact words of Internet searches but this time they have built their model on semantic analysis. Sifting a large corpus of online information, they were able to observe that a fall in the stock market is often preceded by widespread Google searches on topics relating to the economy or the financial markets and so conclude that financial vocabulary as search topics may have great predictive potential.

Predicting market movements based on financial keywords

The researchers used historical data from, among other sources, Google between 2004 and 2012, in order to develop an automated means of spotting user search behaviour likely to precede a dip in quoted share prices. They looked at the main financial terms used in Internet searches, extracted 100 different semantic topics from Wikipedia and drew up lists of the 30 most representative words for each topic. Using the publicly available service Google Trends, the Warwick-Boston team then obtained data on the frequency with which Google users in the United States launched a search on each of these terms. This in turn enabled them to work out to what extent the use of these key terms was related to movements on the stock exchange. They found evidence that Internet searches relating to politics or business revealed some concern about the economy, which was often linked to a drop in investor confidence in equities. In other words, a rising search volume on these specific topics tended to precede a stock market fall.

Real-time semantic analysis able to predict stock market levels?

In developing this method of building models of historical Internet search behaviour and its links with real stock market decision-making, the Warwick-Boston researchers are seeking to explore how far these behaviours can be used as harbingers of future stock exchange movements. In the longer term, their aim is to be able to anticipate the kind of decision-maker behaviour which entails a general dip in share prices, as a means of working out optimal investment strategies. It remains to be seen however whether real-time trend analysis proves to be as effective a predictor of near-future financial behaviour as the a posteriori observations have been for past stock market fluctuations.

By Lucie Frontière