Why are some U.S. cities successful, while others are not? Empirical evidence from machine learning

The U.S. population has become increasingly concentrated in large metropolitan areas. However, there are striking differences in between the performances of big cities: some of them have been very successful and have been able to pull away from the rest, while others have stagnated or even declined. The main objective of this paper is to characterize U.S. metropolitan areas according to their labor-market performance: which metropolitan areas are struggling and falling behind? Which ones are flourishing? Which ones are staying resilient by adapting to shocks? We rely on an unsupervised machine learning technique called Hierarchical Agglomerative Clustering (HAC) to conduct this empirical investigation. The data comes from a number of sources including the new Job-to-Job (J2J) flows dataset from the Census Bureau, which reports the near universe of job movements in and out of employment at the metropolitan level. We characterize the fate of metropolitan areas by tracking their job mobility rate, unemployment rate, income growth, population increase, net change in job-to-job mobility and GDP growth. Our results indicate that the 372 metropolitan areas under examination can be categorized into four statistically distinct groups: booming areas (67), prosperous mega metropolitan areas (99), resilient areas (149) and distressed metropolitan areas (57). The results show that areas that are doing well are predominantly located in the south and the west. The main features of their success have revolved around embracing digital technologies, adopting local regulations friendly to job mobility and business creation, avoiding strict rules on land-use and housing market, and improving the wellbeing of the city’s population. These results highlight that cities adopting well-targeted policies can accelerate the return to growth after a shock.