We investigate the effect of job loss and unemployment benefits on criminal behavior, exploiting individual-level data on the universe of workers and criminal cases in Brazil over the 2009-2017 period. We match workers displaced upon plausibly exogenous mass layoffs with observationally-equivalent control groups to identify dynamic treatment effects of job loss while allowing for treatment effect heterogeneity. In our preferred specification, the probability of criminal prosecution increases by 23% upon job loss and remains approximately constant during the following years. Our unusually large dataset allows us to precisely estimate increases in almost all types of crimes - including offenses with no economic motivation - as well as spillover effects on other household members. The estimated effects remain robust when restricting to arrests "in flagrante", which are less subject to differential reporting by employment status. We then evaluate the mitigating effect of unemployment benefits leveraging on discontinuous changes in eligibility. Regression discontinuity estimates suggest that unemployment benefits covering 3 to 5 months after displacement completely offset potential crime increases upon job loss, especially for liquidity-constrained individuals, although this effect completely vanishes upon benefit expiration. Our findings point at liquidity constraints and psychological stress as main drivers of criminal behavior upon job loss, while substitution between time on the job and leisure does not seem to play an important role.