Web search queries can predict stock market volumes.
In class we studied internet searches and how Google utilized a PageRank algorithm to return search results based on how many other web pages link to or are linked from another web page. Through our homework, we also learned of how it is possible to intentionally “game” or “bomb” Google’s PageRank algorithm to raise the rank of a webpage, and in turn increase its visibility in searches.
Now imagine if it were possible to “game” or “bomb” the stock market using information from web searches. This would be rather difficult to achieve due to the extreme, real time fluctuations of the stock market and the difficulty in filtering out the signals from the noise, but one of the first steps would be to find a positive correlation between the stock market and web searches.
One research study did find a positive correlation between query volumes submitted to various internet search engines and the trading volumes of stocks. This research study specifically utilized a year long interval of Yahoo! web search queries and trading volumes of the NASDAQ-100 stock market, or the 100 largest non-financial companies traded on the NASDAQ. The study made efforts to filter out noise from data, and disregarded correlations that coincidentally resulted from extremely common words such as “fast” and “life.” Overall, the study found an average correlation coefficient around 0.3 between web searches and the trading volumes of various companies in the NASDAQ 100 set.
The study additionally statistically validated its model with a permutation test, that is, they compared the significance of their results against pure randomness, and the overall result was that for three quarters of the stocks, the correlation between query search volumes and trading stock volumes was statistically significant.
The origins of these results can be hard to determine, since there is no solid mechanism of cause and effect for these positive correlations or reasons why web search queries can predict stock trading volume, but the applications are plentiful. The authors of the study themselves say “we think that this information can be effectively used in order to detect early signs of financial distress.”
Rererence: http://arxiv.org/abs/1110.4784