The urban density premium across establishments

We use longitudinal establishment data to estimate the urban density premium for U.S. establishments, controlling for observed establishment characteristics and dynamic establishment behavior. Consistent with previous studies, we find an elasticity of average establishment earnings with respect to metropolitan area population of 0.03, controlling for the endogeneity of location and establishment and metropolitan area characteristics. More importantly, we find that the estimated density premium is realized almost entirely at entry and is constant over an establishment's life. We find little evidence that the endogenous entry or exit of establishments can account for any of the estimated density premium. We interpret our results as implying that the returns to agglomeration diffuse within a city through a reallocation channel rather than through an increase in the productivity of existing firms.