kind of information that Google offers through a service called Google Trends, where
weekly time-series (starting January 2004) of the popularity of any given search term are
available for inspection and download. Looking at search terms relevant from the
investment viewpoint has the potential to correctly identify topics capturing private
investors’ attention and thus give clues as to their future actions. The fact that Google
presently accounts for around 70% of global searches certainly adds weight to this
hypothesis. Da, Engelberg, and Gao (2009) give essentially the same argument and
show how Google Trends can be relevant on the individual stock level. Using Russell
3000 as the universe, they report a statistically significant relationship between the
increase in the search frequency for a stock ticker symbol and the subsequent increase
in private buy orders submitted for that stock. Furthermore, they show how this
contributes to large first-day returns and long-run underperformance of IPO stocks.
Their study is an important step towards documenting the merits of Google Trends in
capturing private investor demand and we build on these findings to illustrate the
resulting market impact.
Instead of focusing on individual stocks we take a different approach based on themes
(or keywords) related to the macroeconomy. We argue that increased interest in those
themes reflects the uncertainty of private investors concerning the macroeconomic out-
look, which might induce increased trading on their part. Correspondingly, to measure
the financial impact we look at the returns, volatility, and implied volatility of the most
popular US index, the S&P 500. We chose to concentrate on three themes: ‘‘recession’’,
‘‘oil price’’, and ‘‘inflation’’ for the period from January 2004 to September 2009. We
decided to concentrate on searches originating in the US only, given the considerable
home bias, characteristic for private investors worldwide. Google Trends values are
calculated as an index and the user can choose between fixed and relative scaling. The
first approach applies the average of search traffic in a fixed time period (generally
January 2004) as a reference value, while otherwise the average for the whole specified
time period is used. While this might seem like a technicality, it gains importance when
applying Google Trends to backtesting. In this kind of setup one has to be especially
careful to clean out any information one could not have had in the past, a problem also
known as filtration. However, downloading one year of Google Trends data with
relative scaling implies knowing the average for the whole year also throughout the
year, which is logically inconsistent. We therefore use fixed scaling in this analysis.
Another controversy, which Da, Engelberg, and Gao (2009) have to deal with is
whether the searches they analyze are indeed linked to investment intentions, as opposed
to looking to buy the company’s products for instance and they argue that searching for
a company ticker rather than its name is strong enough an indication.
We claim that this is not an important issue for us because of the high-level focus
of our study. According to an ICI (Investment Company Institute) report, half of
American households owned stocks in the year 2005, either directly or through mutual
funds. Therefore, greater uncertainty about macro-themes among the general public is
likely to find its way through to the stock market. This argument is further re-enforced
by the fact that we concentrate only on big moves in search interest.
In methodological terms our analysis belongs to the event study type, pioneered for
the stock market by Brown and Warner (1985). Accordingly, we define an event as a net
weekly change in the Google Trends score, which falls in the top 5% of largest changes
up to date (consider again the filtration problem). To establish at least some history, we
Volatility asymmetry, news, and private investors 263