You are using an outdated browser.
Please upgrade your browser
and improve your visit to our site.
Skip Navigation

Google As Unemployment Predictor

What if, through diligent study of search trends, you could predict what happens in the world tomorrow, next week, or next month?

Well, that'll probably never happen, but you might be able to outdo some of the professional forecasters on Wall Street as far as predicting the direction of some important economic indicators.

The idea, featured in a recent paper by Nikolaos Askitas and Klaus Zimmermann of the German-based Institute for the Study of Labor, is simple and clever enough: Askitas and Zimmermann tracked search volume in Germany for phrases that might indicate someone has just lost their job. They used that data to then predict what would happen to the official unemployment rate, which is released in Germany two weeks after the month for which it refers to. The following chart shows how the predictions of one of their models matched up with the official data:

(Look at page 18 of the paper for a larger version.)

The strong correlations here mean that search volume could assist in real-time analysis of the economy. In the case of Germany, Askitas and Zimmermann were able to produce good predictions two weeks before the official data was released. Though when the economy is going through rapid changes, as it did back in the fall, it's unclear how this system would hold up.

The researchers only looked at data from Germany, but a cursory examination of weekly claims data suggest the same method could work in the U.S. I compared the weekly change in seasonally-adjusted initial unemployment claims with the U.S. volume for searches that had the words "unemployment" and "office" in them. It turns out that between 2004 and 2007, the change in initials claims and search volume were in the same direction about 70 percent of the time. (In other words, when search volume for the keywords increased, there was a 70 percent chance that initial claims also increased.) Here is a chart comparing the two:

--Zubin Jelveh