We estimate the causal real economic effects of a randomized anti- corruption crackdown on local governments in Brazil using rich micro-data on corruption and firms. After anti-corruption audits, municipalities experience an in- crease in the number of firms concentrated in sectors most dependent on government relationships. Through the estimation of geographic spillovers and additional tests, we show that audits operate via both a direct detection effect as well as through indirect deterrence channels. Politically connected firms suffer after the audits. Our estimates indicate the anti-corruption program generates significant local multipliers which are consistent with the presence of a large corruption tax on government-dependent firms.