Baidu found China’s “ghost cities,” but it is keeping their locations mostly a secret

China's Transition
China's Transition

Researchers from Peking University and Baidu, the largest search engine in China by far, have drawn on Baidu’s very big data to find the ghostliest cities in the country.

A ghost city, put simply, is a place with far more residential buildings than people. And thanks to a post-global financial crisis construction frenzy—in 2011 and 2012 alone, the country used more cement than the US did in the entire 20th century (paywall)—China has a lot of these.

Based simply on rates of high vacancy, the Baidu/Peking University team found that they mostly appear in sections of mid-sized cities, particularly in the country’s eastern provinces. They identified 50 ghost cities with this method, mostly second- and third-tier cities in Eastern China, but refused to share the names of all these cities, or their vacancy rates, saying that they are “very sensitive” and “can even affect the sale of real estate.”

They did map twenty of them, however, which is a step further than many analysts or economists have gone:

baidu ghost map colorcorrected
20 cities with areas of unusually high vacancy. ("Ghost Cities Analysis Based on Positioning Data in China," Chi et al. 2015)

Since the late 2000s, media have seized upon Chinese ghost cities as a symbol of panglossian excess that results from faith in the “China miracle,” limited investment options, and local governments’ credit-fueled will to build. But it’s harder to pick out ghost cities than you might imagine.

Vacancy should be helpful—if accurate numbers were available. Since the Chinese government declines to publish vacancy rates, some reporters and hedge fund managers have taken to counting apartments with lights off at night. Others calculate that cities with fewer than 5,000 people per square kilometer of residential development make the ghastly cutoff (link in Chinese).

But Baidu’s team clearly doesn’t think much of such approaches. Because these figures don’t include the precise locations of unoccupied housing, they “can only reflect the average level of ‘ghost’ in each city, not to mention finding out the reasons behind ‘ghost cities.'”

To pinpoint a place’s discrete “level of ghost,” the researchers first filtered Baidu Maps data for all of China to zero in on residential areas. Next, they counted the number of people that used Baidu’s search engine in 100-meter square chunks of these residential areas, using their IP address or other location data. The researchers considered an area with less than one-fourth the number of Baidu users than would normally be expected to have a “high vacancy rate,” suggesting a ghost city.

The conveniently round number of 50 they found suggests that a few more have gone unmentioned, as well.

Here are the vacant housing areas for nine of the cities they identified:

nine ghost towns china
Areas of vacant housing in A. Rushan B. Dongsheng C. Kangbashi D. Binhai E. Rugao F. Dongying G. Xinghua H. Yijinhuoluoqi I. Dongling. ("Ghost Cities Analysis Based on Positioning Data in China," Chi et al. 2015)

The dataset used to find ghost cities is huge. It contained a staggering 770 million Baidu users, from September 2014 to April 2015, allowing for unprecedented accuracy in this macro picture of Chinese housing. The data are not perfect, though: Baidu users are disproportionately young and rich, and the identification of residential areas is not perfect, so the research probably left several ghost cities unaccounted for.

The researchers are careful to point out that not all 50 of these are prototypical ghost cities, full of nothing but concrete all year round. Some, they point out, may be tourist destinations where residential areas are largely vacant except during holidays and weekends. They looked at how the population changed throughout the course of the year in two cities: Kangbashi—a.k.a. Ordos, is probably China’s most famous ghost city—and Rushan in Shandong province. The latter looks much more like a tourist spot than a ghost city, because the population jumps during the Chinese New Year holiday.

baidu map activity colorcorrected
User activity in Kangbashi has follows a clear daily cycle; Rushan’s spikes during the Chinese New Year Holiday. ("Ghost Cities Analysis Based on Positioning Data in China," Chi et al. 2015)

That’s some pretty impressive ghost-hunting—for the cities Baidu’s team highlighted, anyway.

home our picks popular latest obsessions search