Past studies have argued that housing is an important driver of business cycles. Housing markets, however, are highly localized, while business cycles are often measured at the national level. We model a national housing cycle using a panel of cities while also allowing for idiosyncratic departures from the national cycle. These departures occur for clusters of cities that experience simultaneous idiosyncratic housing recessions. We estimate the clustered Markov-switching model proposed in Hamilton and Owyang (2012) using city-level building permits data, a series commonly used at the national level as a business cycle indicator. We find that cities do not form housing regions in the traditional, geographic sense. Instead, similarities in factors affecting the demand for housing (such as the average winter temperature and the unemployment rate) appear to be more important determinants of cyclical comovements than similarities in factors affecting the supply for housing (such as housing density and geographic constraints in the availability of developable land).