At the end of 2017, Google accounted for 75% of all search queries globally, making it the market leader. Analysis from digital marketing specialist Smart Insights suggests approximately 3.5 billion searches are performed daily on Google every day.
Google makes the Search Volume Index (SVI) of search terms public via Google Trends. Historical data is available starting from 2004. As soon as you open Google Trends, it greets you with an invitation to explore what the world is searching.’ In fact, Google Trends is an online search tool that allows anyone to see how often specific keywords, subjects and phrases have been queried over a specific period. For example, the top searches in 2017 were Hurricane Irma, Bitcoin and Meghan Markle.
The fact that many of our activities now leave a digital trace (credit card transactions, web searches, e-commerce, GPS) is creating new datasets and new opportunities for researchers to study the world we live in.
There is a body of research that shows how search engine traffic, such as the number of requests submitted by users to search engines, can be used to track and, in some cases, anticipate the dynamics of social phenomena.
One classic example is a paper published by the scientific journal Nature in 2009 titled ‘Detecting Influenza Epidemics Using Search Engine Query Data’. The authors demonstrated that searches for 45 keywords related to influenza and its treatment allow analysts to anticipate by one to two weeks the actual spread of infection as measured by official data on contagion collected by health care agencies. Is it possible to use a similar approach to provide insights on the dynamics of financial markets?
Testing the theory
H. Eugene Stanley, a physicist from Boston University, and behavioral economists Helen Susannah Moat and Tobias Preis from Warwick University attempted to answer this question in their study ‘Quantifying Trading Behavior in Financial Markets Using Google Trends,’ which was published by Nature in 2013.
The team started with a set of 98 search terms and tracked how search volumes on those terms varied between 2004 and 2011.
One insight from the analysis was that searches for the most finance-focused terms such as ‘stocks,’ ‘unemployment’ and ‘revenue’ fell before rises in the market average (proxied by the Dow Jones index), whereas when those terms were searched for more often, the index tended to fall in subsequent weeks.
In particular, the paper focused its results on the term ‘debt,’ as there is an obvious semantic connection to the most recent financial crisis, and it was the term that performed best in their analyses.
The authors investigated a hypothetical investment strategy: buying stocks in weeks when searches for the term ‘debt’ fell, and selling them when volumes rose. Excluding transaction costs, the strategy would have risen 326%, while a simple buy-and-hold strategy in the DJIA would have yielded a profit of 16%, they found.
In 2014, the same authors returned to the subject in a new paper: ‘Quantifying the Semantics of Search Behavior Before Stock Market Moves.’ This time they were interested in understanding financial crises, in particular the earlier stages of this process, when traders may gather information to determine what the consequences of various actions may be.
One of the differences between the two papers is that in the second paper the authors used more sophisticated techniques to choose which words to search. In the second paper, they used a technique from computational linguistics called Latent Dirichlet Allocation (LDA) and applied it to a large online corpus: Wikipedia. LDA is used to identify lists of words constituting semantic topics within a corpus.
Using this technique, they arrived at a list of 55 topics. These topics and their search trends were studied in connection to market moves. Two topics stood out in terms of their ability to predict financial market turmoil: politics and business.
It works, up to a point
Overall, the paper found that increases in searches for information about political and business issues tended to be followed by stock market falls. One interpretation of this is that increases in searches around these topics may constitute early signs of concern about the state of the economy, either from investors themselves, or from society as a whole. However, the authors found that the strength of this relationship has diminished in recent years.
This perhaps reflects increasing incorporation of big data into automated trading strategies, as well as the potential need for more advanced strategies to fully exploit online data in financial trading.
Elisabetta Basilico is a quant investment expert and consultant who specializes in ‘turning academic insights into investment strategies.’ Follow her at: academicinsightsoninvesting.com and on Twitter: @ebasilico.
Site Search 360 Reports