Varian: Google Trends Predicts the Present

If you can predict the future better than other people, you will soon become rich.

If you can know better than other people what is happening right now, that is almost as good. After all, if no one else finds out the truth until a month after the fact, the present might as well be the future — nobody knows it.

For instance, many economic indicators move markets: reports about the unemployment rate, crude oil inventories, etc. All of these are lagging indicators. They report information well after the actual events have occurred. If you can predict the reports ahead of time (and you know how the markets are likely to react to the number in the indicator), you can front-run the market.

I’ve always thought that traffic delays might be a good predictor of current economic activity. If the roads are clogged at rush hour, that is a good sign for economic activity. Busier highways in late morning and early afternoon is probably a bad sign. I’ve never managed to track down the right data to test this hypothesis. It was just too much work.

As with so many other things, Google provides a better way to solve the problem. Hal Varian‘s excellent piece (summarized on the Google research blog, through which you can get the full report) explores how you can substantially improve the predictive ability of forecasting models by using information in Google Trends. It could hardly be easier — certainly much easier than trying to monitor traffic data from 200 different U.S. cities.

Of course, the fact that it is easy is bad news for anyone who wants to profit from it. If things are too easy, everyone can build it into their models, arbitraging away the value.


Gabriel

Sure it's bad news for people who want to profit if it's easy, however it's GREAT for society as a whole.

Eric M. Jones

The process of making things provides many leading indicators: The purchase of land for production facilities, the filing of patents, testing prototypes, the design of tooling--all these actions lead the making of profits by months or years.

For those interested in predictions of any sort would do themselves a favor by finding a dog-eared copy of "the Book of Predictions: http://www.vanityfair.com/online/daily/2008/06/the-book-of-pre.html
This is one of the most valuable books I have. It says for example (In 1979-1980!), Andrew M. Greeley predicted: "Before 1990, the present communist government in the Soviet Union will be overthrown either by a violet internal revolution or more likely by a "social democratic" faction within the party. Some of the constituent republics (the Ukraine, for example) will obtain authentic separate status. The Soviet colonies in Eastern Europe will then go the same route."

Amazing.The CIA was predicting communist world domination at the time. Ask this guy about the economy please.

Read more...

Michael F. Martin

Google is very canny for its decision to make so much of that information available to the public. But nobody has access to it like Google. Google can actually measure the demand curve as it fluctuates in real time. (Neoclassical economists who think the demand curve does not fluctuate in time should not bother replying.)

Is Google already trading on this information? I'd like to know. Would there be antitrust implications if it did? Is that why Google is making so much info available?

Michael Bruns

If it's their data that they gather themselves there's no reason it would be insider-trading.

Robert King

Of course, front-running the market in this manner would require two separate assumptions, right?

(1) The data set being examined is approximately equivalent to the data set being released as an economic indicator.

(2) The data set/economic indicators have the desired effect on the market.

I think the first assumption speaks for itself. As for the second assumption, let's look at housing starts. I remember a friend of mine in the housing industry telling me that when housing starts picked up to 5% for two consecutive months, we would know the housing crisis was over. Unless, of course, the market received the news of additional housing to be a sign of increasing the housing glut as opposed to catching up to increased demand. :-)

misterb

OK, Huffpo says that men's underwear (at least according to Greenspan) is a reliable predictor of economic activity

Google trends show men's underwear searches trending down from 2005, but that the decline has stopped and perhaps an uptick is happening.

There you go, recession solved!

michael

if you get a way to precisely predict, will you share with us,of course you won't.
if you purposely present a way to predict,and you have power to make others to predict that way, then you get it

anon

This is certainly an interesting topic, and here's an interesting discussion of the extent to which searches help.

http://messymatters.com/2009/03/21/the-future-is-yesterday/

Chris S

This twigs as related to an article in the Toronto Star yesterday...

http://www.thestar.com/News/Insight/article/617149

... and the earlier trigger in The Economist...

http://www.economist.com/finance/displaystory.cfm?story_id=13415223

... where it's mentioned that you can estimate the shadow economy by comparing the measured economy to electricity consumption. The shadow economy might not report, but it still uses power.

I wonder if Google has search term usage that corelates to various types of economic activity?

On a similar note, I once recall someone from Google noting that when a local power blackout happens, it takes 10 to 15 minutes to get the very first news report - but the very first web searches start arriving in about 15 seconds.

Joe Smith

There are people who have real time data on the economy.

Any issuer of credit cards knows in real time what consumer behavior is.

Any large national retailer (say Walmart) also has the means to know in real time what consumers are doing.

A bank with commercial clients knows how much money the businesses are depositing and paying out in real time and they know what types of businesses are experiencing what impacts.

econobiker

"On a similar note, I once recall someone from Google noting that when a local power blackout happens, it takes 10 to 15 minutes to get the very first news report - but the very first web searches start arriving in about 15 seconds."

Makes sense with battery backups and laptops and cable enabled internet. And who would care most about a power outage but someone using technology...

earthgecko

Actually the dropping motto appears to be an April Fool's joke... great the Internet you can still be a fool on 1 May!! However, joking aside they have seemingly removed it from their company site. Still looking.. will update.

earthgecko

Confirmed. Google have not dropped their informal "Don't be evil" motto.

http://investor.google.com/conduct.html

I am an April fool :) Confidence is restored ;)

earthgecko

Maybe too long? My post was too long.

That is a very interesting thought Steve, regarding traffic volume as an indicator of economic activity. I have never heard that before. I bet you are right, on most days, when accidents and breakdowns are not greater than average. Actually I bet that the number of accidents would be a fairly good indicator of the volume of traffic. You can see it now on morning traffic shows... "Hey the current economic indicator this morning is 2.3" :)

Nice idea, someone will have the traffic data set, however the point is.... Google.

earthgecko

In response to Joe Smith's comment.
"There are people who have real time data on the economy... Any issuer of credit cards knows in real time what consumer behavior is."

So do you Joe, we all do. twitter, Google, stock market realtime values, etc. However there is no organisation on the planet that matches Google when it comes to being able to process and analyse data. That gives Google the edge over everyone.

jamie andrei

Pervasive comptuing will lead the for this. the next step will be real time, dynamic relationship montitoring from RFID sources etc. Could be the next Bloomberg revenue model. Subscribe to our real time RFID data feed for forecasting and modelling. Kind of like the Twitter style, real time conversation search an metrics available now. A mix of long, short and real time would give some good compartitive tools.