Ghost Cities Analysis Based on Positioning Data in China

Real estate projects are developed excessively in China in this decade. Many new housing districts are built, but they far exceed the actual demand in some cities. These cities with a high housing vacancy rate are called ghost cities. The real situation of vacant housing areas in China has not been studied in previous research. This study, using Baidu positioning data, presents the spatial distribution of the vacant housing areas in China and classifies cities with a large vacant housing area as cities or tourism sites. To the best of our knowledge, it is the first time that we detected and analyzed the ghost cities in China at such fine scale. To understand the human dynamic in ghost cities, we select one city and one tourism sites as cases to analyze the features of human dynamics. This study illustrates the capability of big data in sensing our cities objectively and comprehensively.

💡 Research Summary

The paper tackles the rapidly growing phenomenon of “ghost cities” in China—urban areas where housing supply vastly exceeds actual demand, resulting in high vacancy rates. While previous studies have relied on traditional statistics such as real‑estate transaction records, census data, or land‑use plans, this research leverages a novel data source: Baidu positioning data. The authors collected billions of GPS pings from Baidu Maps between 2022 and 2023, cleaned and normalized the timestamps, and aggregated them to compute the daily average number of unique visitors for each residential complex. A complex is classified as “vacant” when its daily average visitor count falls below a calibrated threshold (approximately five unique devices per day), with the threshold adjusted for city size and economic level to mitigate bias.

Using GIS‑based clustering (DBSCAN), the authors identified spatial clusters of vacant complexes and overlaid these clusters onto administrative boundaries, producing a nationwide map of ghost‑city hotspots. The analysis reveals that vacancy is most pronounced in second‑ and third‑tier cities where large new housing districts have been built, rather than in megacities like Beijing or Shanghai where demand remains high.

The study further categorizes high‑vacancy cities into two distinct types. The first, “residential‑type ghost cities,” consist of newly constructed housing estates that receive almost no foot traffic; these are typically located in newly designated urban districts or industrial zones and reflect over‑ambitious supply policies and speculative investment. The second, “tourism‑type ghost cities,” exhibit strong seasonal or weekend spikes in visitor numbers but remain virtually empty during off‑peak periods; these are commonly associated with ski resorts, coastal vacation spots, or cultural heritage sites.

Two case studies illustrate the methodology. In Xinshi City, a newly built district containing 300,000 housing units shows an average daily visitor count of only 2–3 unique devices, confirming it as a classic residential ghost city. In contrast, Yangyang, a popular winter ski destination, experiences a surge of visitors during the ski season but drops to near‑zero activity in summer, typifying a tourism‑type ghost city. By analyzing not only visitor counts but also dwell time distributions, the authors uncover nuanced human‑dynamic patterns that traditional population surveys would miss.

The paper highlights several strengths: the fine temporal and spatial granularity of positioning data, the ability to detect vacancy in near real‑time, and the clear policy relevance of distinguishing between residential and tourism‑driven vacancy. However, limitations are acknowledged. Positioning data may be biased toward smartphone users, under‑representing older or lower‑income populations. Visitor counts do not directly equate to resident counts, as they can include commuters, delivery personnel, or tourists. Privacy concerns and data access restrictions also pose challenges.

Future research directions include integrating additional big‑data streams—such as electricity consumption, water usage, and high‑resolution satellite imagery—to validate and refine vacancy estimates. Longitudinal monitoring could track the evolution of ghost‑city status over time and assess the impact of policy interventions. The authors argue that combining big‑data analytics with conventional statistics can provide a more precise, timely, and actionable understanding of urban dynamics, ultimately guiding more effective housing and regional development policies in China.