Serial Bus

One post per week

Predicting Flu, Recession and More….

leave a comment »

Today, I spoke with Nitesh, an old friend, after a long time. When New York catches a cold, Bangalore is one of the first cities to sneeze. So, being based there, Nitesh is witnessing the unraveling of the global economy from close quarters. As we got talking, one of the outrages that we shared was why nobody was able to predict its onslaught. While some players in the economy have a vested interest in not blowing the whistle, the same is not true of the numerous state and non-state agencies (starting with Uncle Ben) whose professed goal is to tell as they see it. Even Alan Greenspan seemed befuddled recently with his own lack of clairvoyance. Beyond what we have already heard and read, I do not have any insights into why the wise folks did not see this coming. Rather, I want to point out how Google has shown us an interesting way of doing what armies of analysts failed to do i.e. predicting recession.

Around the outbreak of the flu season in the US last year, Google unveiled a fascinating avatar of its Trends utility – Flu Trends. Normally, the CDC (Centre for Disease Control & Prevention) issues announcements about advances of the flu virus every year. This information is collected from the various medical facilities that report the increased incidence of patients suffering from flu. Given the somewhat post-facto nature of this information, by the time CDC reports the spread of flu virus in say, the Mid-Atlantic, lots of people have already contracted it and suffered.

Google’s brainwave is very simple and elegant. Mining through piles of search strings used by its users, Google realized that there is a robust direct co-relation between increased incidence of flu-related searches and its outbreak in any given area. Obviously, a lot of people google their symptoms (and perhaps the word ‘flu’) when they start feeling weird, especially if it’s the season. It does not mean that every person searching thus is going to be sick, but the margin of error (or the standard deviation) is very small. Here is the beauty of this – using this data, Google is able to predict the flu wave 2 weeks before CDC gets the wind through its traditional channels. This has significant implications. By collaborating with Google, local governments and CDC can raise their levels of preparedness closer to the outbreak and contain it before it spreads to the larger population.

Now, I think this technique can be applied to other events that we would like to predict and remedy. You are right, I am talking recession.

A recession is usually defined as two consecutive quarters of decline in real GDP. Real GDP is a direct indicator of economic activity in a country and when this activity is not keeping up, what kind of searches would you expect people to perform on Google? Some usual suspects are ‘jobs’, ‘how to save’, ‘lay-off’, ‘low prices’, ‘foreclosure’, ‘credit rating’, ‘default’, ‘grocery discounts’ etc. Moving away from individuals for a while, there are some other types of searches that small businesses might perform on items like (low) sales, (decreased) credit line, (excess) inventory etc. If Google Trends is able to plot this data against a timeline, I am very certain that it will match closely with the data collected by the National Bureau of Economic Research (NBER). If this co-relation can be established, Google can forecast any downturn much before the NBER suspects anything. Because of data revisions and lags in data availability, recessions are often announced retroactively by the NBER.

Take the current recession. Late in 2008, the NBER announced that economic activity had peaked in December 2007, and that the U.S. economy had entered a recession in December 2007. However, at the time of the NBER announcement, the recession was already a year old and already one of the longest recessions since World War II.

Having said that, I wonder what other phenomena can Google Trends fast-track the prediction of? There should be several, some trivial and some significant. For instance, tax software companies can find out how close to the returns deadline do people begin to worry about them – and then, maximize the use of their advertising dollars in that time period. Other possibilities include prediction of election victories and winners of American Idol.

There are only four criteria for Google Trends to have a close co-relation with the actual data: (1) The phenomenon must have a strong human element (sorry, can’t predict earthquakes) (2) It should be mass-based (sorry, can’t predict the next scandal involving Britney Spears) (4) Its ascent should be gradual (a good contrasting example, anyone?) and (3) Google should be the primary search engine accessible and used by majority of the population (sorry, can’t predict stuff in Somalia).

So, if you feel like Nostradamus some day and the object of your curiosity matches the above criteria, well, help is just a click away!!

Advertisements

Written by serialbus

March 22, 2009 at 8:12 pm

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: