By Daniel Akst
Wondering if the market is in for a tumble? If you find yourself
Googling a lot of political and financial terms, it may well
be.
In the past, trends in Google searching have been shown to
predict flu outbreaks, unemployment rates and the success of movies
at the box office. There was even evidence that financially
oriented searches--of Google and Wikipedia--could predict stock
market movements.
The pattern was simple: Before stocks moved lower, there was an
uptick in searches of finance-related terms. Now researchers from
Boston University and the University of Warwick, in England, are
reporting that stepped-up searching for terms relating to politics
also points to a lower market.
Building on their previous work, the scientists used the
techniques of computational linguistics to group all the words in
Wikipedia into topics. Then, using Google Trends, a publicly
available service, they determined how often salient keywords
within each topic area were searched from 2004 to 2012. The result?
Increased searching of finance or political terms predicted falling
stocks.
The scientists used their keyword searches to make hypothetical
trades based on historical data for the S&P 500 index. For each
keyword topic, they bought or sold the index weekly depending on
whether searches were rising or falling, comparing their results to
a strategy of buying and selling randomly each week. The median
return for trading based on a collection of politics keywords was
38% above the random strategy. For business-oriented keywords, it
was 28%. "Crucially," the scientists write, "we find no robust link
between stock-market moves and search-engine queries for a wide
range of further semantic topics."
They also found that the power of Google searches to predict
stock market movements has diminished in recent years.
"Quantifying the Semantics of Search Behavior Before Stock
Market Moves, " Chester Curme, Tobias Preis, H. Eugene Stanley and
Helen Susannah Moat, Proceedings of the National Academy of
Sciences (July 28)
Subscribe to WSJ: http://online.wsj.com?mod=djnwires