Google searches may contain useful information for forecasting near-term S&P
500 Index returns, assuming that such terms as “market crash,”
“market rally,” and “bear market” proxy effectively
for investor sentiment. Conversely, near-term past market performance may affect
the same sentiment measures, possibly consistent with a “gambler’s
fallacy” or “hot hands” effect.
The authors analyze Google Search Volume Index (GSVI) data for search frequencies
related to four search terms: “market crash,” “market
rally,” “bear market,” and “bull market.” They
hypothesize that changes in relative search volume proxy for immediate market
sentiment because they represent where average investors might be focusing their
attention. They predict that negative near-term returns will follow an increase in
the volume of negative search terms (crash and bear) and that positive returns will
follow an increase in the volume of positive search terms (rally and bull). Their
findings indicate predictive information in three terms (crash, bear, and rally) but
no significant information in one term (bull).
How Is This Research Useful to Practitioners?
If search term volumes contain reliable indicators of current investor sentiment,
then traders could monitor such volumes and perhaps gain an edge over traditional
fundamental or technical analysis when timing a trade. A persistent indicator would
also find its way into the academic toolbox, and we should expect that such results
would drive other studies that explore additional search terms.
If Google controls the search term volume data, it could sell this information to the
highest bidder, possibly making it cost-prohibitive for smaller investors. And if
persistent advantages are found, traders could compete them away over time so that
immediate market sentiment becomes the new standard used by all. Or perhaps robots
would be designed to flood the Google search engine with false search terms in a new
form of market manipulation.
To a great extent, the validity of the authors’ findings depends on whether
search term volumes correspond to some sort of observed behavioral reality. A
correlation between search term volume and subsequent or prior market performance
appears to exist, but it is critical to understand why the original search behavior
exists in order to reach a valid conclusion that pushes the findings beyond the
How Did the Authors Conduct This Research?
Using GSVI data from January 2004 through February 2011, the authors compare weekly
lagged differences in search volumes for the specified search terms against the
average search volume for the same search terms over different review periods.
Ordinary least-squares regressions are used to identify relationships between future
and past S&P 500 Index total returns and the changes in search volumes.
Robustness tests involve vector autoregression and Granger causality models as well
as comparisons with changes in the American Association of Individual
Investors’ individual investor sentiment measure, a much-studied control
variable. Notably, the period of study includes the Great Recession and two
recoveries, which the authors believe provides them with different test
environments. They note their inability to test the term “market rally”
over one early period because of a lack of data. Their results do not indicate
multicollinearity in the regressions.
Among the four search terms, the strongest directional relationships the authors find
are for the terms “market crash” or “market rally.”
(Results for “bear market” are mixed, and “bull market”
has no significance.) The authors interpret these results as evidence that current
investor attention relates to prior sentiment measures and can translate into future
trading activity, especially when attention turns negative. But near-term past
S&P 500 returns also seem to influence current market attention, although
somewhat differently. After negative returns, searches for both “market
crash” and “market rally” increase, which is perhaps explainable
by the “hot hands” effect (i.e., attention follows the observed trend)
in the first case and “gambler’s fallacy” (i.e., attention
anticipates reversals) in the second.
The authors expand earlier research related to immediate investor sentiment by
analyzing Google search terms. Intuitively, investor sentiment should affect
internet searches, and the authors demonstrate a relationship between certain terms
and market movements. But they do not establish why they expect people to search for
the terms they choose in the first place. It seems odd to call out these directional
terms as indicative of market sentiment when investors might instead search for
“commentary,” “recap,” “business news,” and
the like. To add more weight to this study’s results, the authors should
demonstrate why they expect specific search terms.