Estimating the Intra-Urban Wage Premium

We estimate the intra-urban wage premium and its attenuation with distance from a balanced panel of workers in the São Paulo Metropolitan Region in the period 2002–2014, dispersed in a fine grid of 1km x 1km cells. We do not determine exogenously central business centers (CBD) or sub-centers (SBD), but estimate the wage premium for each cell and for two layers of surrounding cells. Based on the wage premium estimated in these three layers, we were able to estimate the attenuation effect. This way of estimating the wage premium and its attenuation in space is novel to the literature. We estimate POLS and Fixed Effects specifications to illustrate how the sorting of workers and firms is powerful in biasing the estimated coefficients. Since sorting might be present even after the introduction of fixed effects of worker, firm, cell, sector-cell and matches firm-worker-cell, we run a 2SLS-IV model. We instrumentalize employment in the cells by the additional constructed area for business and housing purposes, lagged by 7 years. Even after controlling for those time-constant effects, our instrumental variable estimation provides evidence of an intra-urban wage premium and its non-isotropic agglomeration effects. The estimated premium is larger by multiples of the POLS or Fixed Effects estimations. The magnitude of the effects is economically relevant: the introduction of 100,000 additional workers in the inner cell produces a 1.78% increase in wages in that cell. The effects on the surrounding cells are negative and much larger than in other forms of estimation that do not correct for endogeneity. The heterogeneity analysis revealed that educated workers benefit the most from the increased interaction possibilities given by concentration and they are less affected when located in the surrounding cells.