July Unemployment Numbers (U.S.) – Calling All Economists

When I’m not crunching the latest numbers on search engines or predicting winners of American Idol, one of my pastimes (and perhaps the reason for my reputation as a data geek) is correlating Hitwise data with economic indicators.

When I started with Hitwise in 2004, I had an idea: If more and more individuals filing for unemployment are doing so online, and I can track that activity via a custom category of the largest state unemployment sites, then I should be able to create a chart of online unemployment activity. Since Hitwise data on market share of visits is available monthly, weekly and daily, and the Department of Labor monthly unemployment numbers are reported monthly with a few days lag, I should be able to predict (at least directionally) unemployment rates in advance of the DOL announcement.

This week, news broke that the U.S. unemployment rose unexpectedly to 4.8%. Based on Hitwise data I anticipated this rise in unemployment over two weeks ago. Here’s the Department of Labor Chart:

Department of Labor Unemployment Rate Chart

Compare that with the Hitwise chart below:Traffic to Unemployment Websites and for keyword unemployment

The blue line is market share of visits for a custom category that I created of the seven largest state unemployment website. The red line is the weekly volume of searches on the term “unemployment.” Correlating these data points to the DOL number is not an exact science. However, over the last two years, I’ve noticed that when both the visits to the unemployment sites and searches on “unemployment” show a directional change (either increasing or decreasing) that change is reflected in the DOL announcement.

This is just a rudimentary demonstration of how we can leverage online behavior to form leading indicators to traditional leading indicators. I’ve been experimenting with some others, such as predicting housing starts, consumer confidence and retail spending to name a few. If anyone has suggestions for additional indicators or ways of refining the unemployment example above, I’d be interested in hearing your thoughts.