This paper exploits time and geographic variation in the adoption of Special Economic Zones in India to assess the direct and spillover effects of the program. We combine geocoded firm-level data and geocoded SEZs using a concentric ring approach, thus creating a novel dataset of firms with their assigned SEZ status. To overcome the selection bias we employ inverse probability weighting with time-varying covariates in a difference-in-differences frame-work. Our analysis yields that conditional on controlling for initial selection, SEZs induced no further productivity gains for within SEZ firms, on average. This is predominantly driven by relatively less productive firms, whereas more productive firms experienced significant productivity gains. However, SEZs created negative externalities for firms in the vicinity which attenuate with distance. Neighbouring domestic firms, large firms, manufacturing firms and non-importer firms are the main losers of the program. Evidence points at the diversion of inputs from non-SEZ to SEZ-firms as a potential mechanism.