Do state borders constrain internal mobility? Evidence from Brazil

Abstract

This paper documents that state borders are a meaningful source of friction in internal labour mobility in Brazil, a large emerging economy with substantial labour-market informality and historically heterogeneous borders. Using geocoded municipality-to-municipality migration and commuting flows from the 2010 Brazilian Census (20.6 million person records, 5565 municipalities), at equivalent distances Brazilians are about 1.8 times as likely to migrate to a same-state municipality as to one across a state border (1.61–1.74 once travel-cost controls are added), and 1.87 times as likely to commute within-state. Three patterns of heterogeneity shed light on the underlying mechanisms: (i) formal-sector workers face a proportionally larger penalty than informal workers (Riemann ratio 2.12 vs. 1.70); compositional reweightings absorb about 45 percent of the raw migration-rate gap but attenuate the formal border-penalty coefficient by only 12–14 percent, leaving a residual consistent with state-scoped institutional channels such as professional licensing, employer concentration and recruitment networks; (ii) colonial-era borders impose larger penalties than recently created ones, a cross-sectional pattern consistent with home-state bias developing gradually; (iii) the penalty is concentrated among individuals living in their birth state (Riemann ratio 12.02) and reverses for interstate migrants (0.45). Fiscal discontinuities in state consumption taxes (ICMS) provide no support for a tax-based mechanism. Headline magnitudes are rate-based Riemann ratios; the conditional Poisson pseudo-maximum likelihood (PPML) border coefficient implies a smaller penalty (exp(0.289) ≈ 1.34), reflecting conditioning and flow-weighting. The results are robust to controls for physical geography (rivers, elevation, terrain ruggedness), paved-road distance and driving time, their joint inclusion, two-way clustering, PPML gravity estimation with origin- and destination-state fixed effects, distance bandwidths up to 200 km, leave-one-out exclusion of the Tocantins border, population-size matching and a permutation placebo test.