Google uses R to predict economic activity

Predicting economic activity is an important business. For example, Wall Street considers Government reports of retail activity and property sales as leading indicators of the economy as a whole, and the Dow Jones Index sometimes rises (or falls!) markedly based on the reported numbers.

One problem: those government reports are released weeks after the fact (the August housing report was released only last Thursday), and even then only with preliminary data to be revised upwards or downwards later. Is there a way to get a jump on these reports and get a measure of economic activity in real time, predicting the outcomes of the official reports?

Google's Chief Economist studied just this question with a research paper (PDF) announced in a Google Research blog post back in April. The key insight is that the volume of Google searches for particular keywords is correlated with related economic indicators. For example, here's a comparison of searches for the keyword "Ford" with actual monthly sales of Ford vehicles:

Varian asks the question: can we use Google Insights data like this to predict the economic variables, even before they are reported?

Seems like you can. Varian used R to fit seasonal autoregressive models to retail sales, automotive sales, home sales, and passenger arrival data, and in each case made better predictions by including Google Trends data as a predictor than without. You can try this yourself by downloading R and using Google Insights for Search to download the trend data -- Varian has helpfully provided the R code for the Ford model in the paper. (Just be sure to log into Google Insights with your Google Account to enable downloading of trends data as a CSV file.)

Now, Google has taken this one step further by introducing Google Domestic Trends, where you can track the ups of downs of searches in 23 categories including air travel, real estate, and retail trade. Will Wall Street use these indices of search activity (and R!) to predict the economic reports before they come out? We'll know it's working when the Housing Report no longer sends the Dow into a tailspin.

Google Research Blog: Predicting the Present with Google Trends (with thanks to Roger Koenker for pointing out the existence of this report)

Comments

[Paul Bivand]

The German IZA (Intitute for the Study of Labour) uses Google econometrics to predict unemployment. This is covered in a paper in Applied Economics Quarterly. The paper is freely available from this address: http://www.atypon-link.com/DH/doi/abs/10.3790/aeq.55.2.107

Not sure if they use R, as no citation for software, but some of the graphics look distinctly ggplot2-ish.

[NotMe]

The Housing Report will send the S&P into a tailspin as long as people believe that it will send the S&P into a tailspin :-)

[sprucesquad]

Thank you for your explanation. There really useful information.

[Air Jordan Shoes]

*People enjoy the characters are not happy.