Hal Varian has a paper out on Google Trends predicting changes in New Claims for Unemployment Benefits. I haven’t actually read it yet but the post announcing it was so exciting that I went ahead and downloaded some data.
I am sure Varian has much more to say but this initial run was just too cool not to post
The blue line is people actual New Claims data, unadjusted for seasonal changes. The red line is the relative number of people searching for “Unemployment Benefits.” There may be a better search query but this is the first thing that occurred to me.
The relationship between the curves is so striking. This is just people searching for the term “unemployment benefits!” This is not adjusted by news items which might effect the number of searches. It also not a more concentrated search like “how to file.” I will read and post more later but this is just too cool!
I have to repeat, the red line has not been adjusted or econometrically fit in any way. This is a simple raw query! Imagine the possibilities!
HT: Econbrowser
3 comments
Comments feed for this article
Tuesday ~ July 28th, 2009 at 1:38 am
teageegeepea
And they laughed at Larry Summers for citing Google search trends as evidence for the health of the economy. I admit I snickered a little bit while accepting it as some Bayesian evidence.
Tuesday ~ July 28th, 2009 at 11:16 am
roland
Excellent and striking. What accounts for the separation beginning in January? The search term lags actual claims? Doesn’t really look like it..
Wednesday ~ July 29th, 2009 at 9:24 am
ao
Klaus ZImmermann and other economists at the Institute for the Study of Labor (IZA) have done further explorations of using google data to forecast unemployment. They published a paper on the methodology in the Applied Economics Quarterly, and have another working paper with an application to German unemployment forecasts . Ungated version is of the former is here http://ftp.iza.org/dp4201.pdf and the latter is here http://ftp.iza.org/sp13.pdf . Unfortunately the second one is in german, an english version exists somewhere (otherwise how did I read it?) but I can’t find it off the top of my head.