First-hand information for everyone
This paper shows that cities, not countries, are the right scale to understand much economic growth. Using satellite night-time lights as a proxy for economic activity, the authors build yearly GDP-like growth paths for 8,808 functional urban areas (FUAs) in 165 countries between 1993 and 2019. When they cluster those full time series, they find 17 persistent groups, or “regimes,” of city growth with different volatility, long-term trends, and responses to shocks.
To find these regimes the researchers first reduced the complexity of each city’s growth path using principal component analysis (PCA), a standard way to keep the main patterns while removing noise. They then grouped cities with similar temporal patterns using k-means clustering and settled on 17 clusters by maximizing a silhouette score, a measure of how well-separated the groups are. The clusters are internally coherent and stable over time, which the authors interpret as evidence for persistent regimes rather than temporary deviations.
The regimes cut across national borders and often split individual countries into multiple types. Some clusters are geographically concentrated, such as a post‑Soviet bloc, while others are dominated by fast state-led industrializers like China. A few regimes are essentially single-country groups, for example Venezuela and Iraq, which the authors say reflect extreme volatility tied to political instability and resource dependence. Regime membership explains an additional 16% of within-country variation in urban growth beyond national effects, suggesting these groupings capture meaningful structural differences.
The paper also builds a directed network of how growth shocks move between regimes, using lagged correlations of regime-level growth rates. They find that shocks tend to travel along lines of structural similarity rather than by physical proximity. Advanced-economy regimes tend to “export” disturbances, while many emerging-economy regimes act as absorbers or amplifiers. The authors argue this pattern challenges standard models that treat countries as independent units and that predict a single global convergence path.
Important caveats are noted. Night-time light intensity is used as a proxy for GDP; the authors cite prior work showing a strong correlation in many settings, but any proxy has limits and can miss aspects of economic activity. The clustering results depend on methodological choices (PCA, k-means, and the criterion for choosing 17 clusters), so regime boundaries are not unique and should be interpreted with caution. The shock-propagation analysis is based on correlations, which do not prove causal transmission mechanisms. Still, the findings point to the value of looking below the national level: cities occupy different “economic niches,” and policy that ignores that spatial and structural diversity may miss where growth actually happens and how crises will spread.