Can Internet Search Queries Help to Predict Stock Market Volatility?

Date01 March 2016
Published date01 March 2016
Can Internet Search Queries Help to
Predict Stock Market Volatility?
Thomas Dimp
University of Tübingen, Germany
Stephan Jank
Frankfurt School of Finance & Management and Centre for Financial Research (CFR), Cologne,
We study the dynamics of stock market volatility and retail investorsattention to the
stock market. The latter is measured by internet search queries related to the
leading stock market index. We nd a strong co-movement of the Dow Jones
realised volatility and the volume of search queries for its name. Furthermore,
search queries Granger-cause volatility: a heightened number of searches today is
followed by an increase in volatility tomorrow. Including search queries in
autoregressive models of realised volatility improves volatility forecasts in-sample,
out-of-sample, for different forecasting horizons, and in particular in high-volatility
Keywords: realised volatility, forecasting, investor behaviour, limited attention,
noise trader, search engine data
JEL classification: G10, G14, G17
1. Introduction
Large stock market movements capture investorsattention. This fact can be seen in
Figure 1 which depicts a strong co-movement between the volatility of the Dow Jones
stock market index and internet search queries for its name. If volatility is low, searches
related to the Dow Jones are generally at their overall average level, but search queries
We thank an anonymous referee, John A. Doukas (the editor), Jeremy Ginsberg, Joachim
Grammig, Isaac Hacamo, Heiko Jacobs, Kajal Lahiri, Terrance Odean, Franziska Peter,
Tomas Reyes, Maik Schmeling, Hal Varian, and Qingwei Wang for helpful comments and
suggestions. We retain responsibility for all remaining errors. Financial support of the
German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) is gratefully
European Financial Management, Vol. 22, No. 2, 2016, 171192
doi: 10.1111/eufm.12058
© 2015 John Wiley & Sons Ltd
surge to levels exceeding their average by multiples during times of market turbulence.
For example, when the volatility of the Dow Jones spiked to an almost record high of over
150% annualised on 10 October 2008, the number of previously submitted searches for
the index had increased to more than eleven times the average.
Internet search queries related to stock market index names can be interpreted as a
measure of retail investorsattention to the stock market (cf. Da et al., 2011). While
professional investors monitor the leading index all the time, retail investors presumably
do not. Once the latter perceive an increased demand for information about the stock
index, they are likely to use the internet via a search engine like Google as a source of
information. In contrast, professional investors most probably, if not certainly, are not
using a search engine to obtain information about the leading stock market index. Thus,
search queries qualify as a good proxy for retail investorsattention to the stock market.
After having searched for the stock market index, some individuals might be inclined to
act and trade immediately or the following day.
Retail investors are often considered to be uninformed noise traders. This notion is
supported by empirical evidence showing that retail investors lose money with their
trading decisions (e.g. Odean, 1998; Grinblatt and Keloharju, 2000). Several behavioural
nance models (e.g. Black, 1986; De Long et al., 1990) predict that trading of uninformed
noise traders can temporarily lead to mis-pricing and increase volatility. Similarly, in the
agent-based models of stock market volatility (e.g. Lux and Marchesi, 1999; Alfarano and
0.02 .04 .06 .08 .1
Realised volatility
2007 2008 2009 2010 2011 2012
Realised volatility of the Dow Jones
0 5 10
Search Volume Index
2007 2008 2009 2010 2011 2012
Search queries for the index name
Fig. 1. Realised volatility and search activity.
This gure presents daily realised volatility of the Dow Jones Industrial Average index (upper
graph) and the volume of Google search queries (lower graph) for its name from 1 July 2006 to 31
December 2011. Search queries are standardised such that the sample average equals one.
© 2015 John Wiley & Sons Ltd
172 Thomas Dimpand Stephan Jank

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT