Traffic congestion keeps rising. In 2014, the German Automobile Club (ADAC) recorded 475,000 traffic jams on German highways, totaling 960,000 kilometers of jammed traffic. Side effects include increased carbon emissions, energy waste, higher transportation and production costs, waste of labor, and delays in product deliveries. A consortium of German car manufacturers estimates the daily economic damage at about 250 million euros.
As the emergence of traffic jams is a complex matter with many potential causes, forecasting traffic jams is a methodologically demanding task. In a new IZA Discussion paper, Nikos Askitas proposes an elegant and parsimonious way to capture expected road congestion before it appears. He utilizes the fact that car drivers tend to inform themselves via internet search about pre-existing traffic conditions, thereby revealing their planned itineraries in advance.
His results show that publicly available Google search intensity data allows to accurately predict 80% of the variation in ADAC traffic jam reports two hours in advance: a 1% increase in searches for the term “stau” (traffic jam) implies a .4% increase in traffic jam reports two hours later.
In a nutshell, the Google searches at 7:00hrs and 16:00hrs predict how bad things will be at the peak of badness at 9:00hrs and 18:00hrs, respectively, netting out fluctuations by time of day and day of week. Geographic information incorporated into Google searches, such as specific highway numbers and regions, provides further information for a higher quality of forecasting.
The author argues that traffic planners would be well advised to take Google searches for traffic congestion into account when developing models for traffic congestion forecasting and prevention. He further highlights the need for more detailed target data, e.g. through GPS information of search engine users.