As the web becomes ever more integral to daily life, researchers are mining web data to predict economic and financial data releases. Three recent academic studies highlight the potential importance of web data to financial market analysis. It is only a matter of time before we see the emergence of a new class of investment research based on web data analysis.
We recently reviewed studies examining the correlation of web search data to financial market data such as mortgage refinancing, foreign exchange rates, stock market volumes, and stock prices. The studies evidence the growing power of the internet, and strongly suggest that financial market participants will be able to extract value from analyzing web data.
A study by Da, Engelberg and Gao published by the Journal of Finance in October 2011 found that Google search data had predictive value for stock prices and IPOs. The study used weekly Search Volume Index data from Google Trends to predict higher prices in stocks favored by retail investors over the succeeding two weeks, followed by price reversals over the subsequent year. The data also predicted first day IPO returns and subsequent price declines. The study was conducted on stocks included in the Russell 3000 in the period from January 2004 to June 2008.
A study by academics associated with Yahoo! Research made available by Cornell University Library in October 2011 found Yahoo! search data was predictive of stock market volumes. The study examined stocks in the NASDAQ-100 index during the period from May 1, 2010 to April 30, 2011. The overall correlation was between 30 – 35%, and in the cases of some stocks as high as 80%. (See graph of the search queries and trading volumes for NVDA below.)
Two economists with the NY Federal Reserve posted a note in January 2012 discussing recent trends in forecasting economic and financial releases using web search data. Included was their own analysis of mortgage finance applications, showing that the search term “mortgage refinance” had predictive value in forecasting future releases of the weekly index of mortgage refinancing. The article also mentions that they found that search terms related to the renminbi had strong predictive value in forecasting the dollar – renminbi forward rate. They attribute this to language barriers and other information barriers relating to mainland China. In contrast, they indicated analysis of search terms relating to gold prices, European sovereign spreads, interbank rates, and implied volatility of equity markets came up dry.
The NY Fed article also contains a good summary of recent studies relating to web data. Da, Engelberg and Gao expect more studies to come: “Search volume is an objective way to reveal and quantify the interests of investors and therefore should have many other potential applications in fi nance.”
We have previously highlighted research firms which try to correlate brand sentiment derived from web data to securities values. We have also noted previous attempts at semantic analysis of Twitter posts and other web content. There is no question we will be seeing more of this type of research in the future.