"Where Sentence Is Served: Entrepreneurship and Socioeconomic Mobility Among Ex-Offenders" (with Vera Rocha), Organization Science, 2026.
Entrepreneurship is often promoted as a pathway to upward mobility, especially for those who face barriers in traditional labor markets. However, whether and when entrepreneurship truly improves the long-term prospects of individuals with criminal records remains poorly understood. This study examines how the type of sentences — community service versus short incarceration — shapes ex-offenders’ employment routes and socioeconomic outcomes. Using three decades of Danish registry data, we compare individuals convicted of similar offenses but given different sentences and follow their trajectories over time. We find that incarceration substantially increases the likelihood that ex-offenders enter entrepreneurship as an alternative to wage work. However, this career path proves costly: compared with those who secure regular employment, individuals who pursue entrepreneurship after release earn persistently lower income and are more likely to reoffend. Mechanism analyses show that transitions into entrepreneurship concentrate among groups facing the steepest hiring barriers—foreign-born individuals, those without vocational training, residents of high-unemployment areas, and offenders convicted of property crimes or serving longer sentences—suggesting that entrepreneurship often reflects “push” rather than “pull” dynamics. Relative to unemployment, entrepreneurship is associated with modestly lower recidivism but similar income penalties. Our findings add nuance to the narrative of entrepreneurship as an emancipatory route for marginalized populations. They highlight the risks of relying on entrepreneurship as a re-integration strategy and underscore the need for organizational and policy interventions that expand access to stable jobs and provide targeted support for people with criminal records.
"Decomposition methods" (with Sergio Firpo), Handbook of Labor, Human Resources and Population Economics, 2023.
This chapter uses the classic Oaxaca-Blinder (OB) decomposition for the mean as its point of departure and then focuses on the last 15 years of decomposition methods that go beyond the mean. The analysis starts from the formal decomposition theory and its link with the treatment effect literature. Then, the OB decomposition is formally derived, and next, the other methods for decomposition are shown. Finally, a link is made between this literature and structural models. In this way, this chapter can be seen as a guide for performing decomposition methods.
"Passing through the supply chain: Implications for market power" (with Carolina Melo and Rodrigo Moita), International Journal of Industrial Organization, 2021.
In this paper, we study the connection between pass-through and market power in the Brazilian Liquefied Petroleum Gas (LPG) industry. We use a state tax shock and apply a difference-in-differences strategy to estimate pass-through –at different levels of the supply chain –and an instrumented difference-in-differences strategy to estimate demand and then feed a theoretical model to make inferences about a conduct parameter –that measures market power. We find an incomplete pass-through at the distribution level, for the supply chain as a whole, and a more-than-complete one at the retail level. Furthermore, we estimate the price elasticity of demand to be greater than one. When we feed a theoretical pass-through model under imperfect competition with these estimates, we obtain a high conduct parameter at the whole supply chain level and an even higher one at the retail level alone. These results show that considering only the whole supply chain pass-through may lead to hasty conclusions about market power. Besides contributing to the empirical literature that connects pass-through with market power, we contribute to ongoing national discussions regarding competitiveness in the LPG industry.
"Sample selection in unconditional quantile models".
We propose constructing a consistent estimator that addresses the problem of sample selection in unconditional quantile models. The proposed approach is based on three steps: (i) estimation of a control function using a logistic distribution regression; (ii) construction of a counterfactual distribution of the latent dependent variable conditional on the previously estimated control function; (iii) application of the recentered influence function (RIF) on the estimated counterfactual distribution and, finally, we run an ordinary least squares regression.
"Challenges and solutions to maintain the validity of the experimental analysis of large social programs in the public sector" (with Ricardo Paes de Barros, Sergio Firpo, Ricardo Henriques, Laura Machado, and Mirela Carvalho).
In this paper, we document the challenges faced and solutions proposed by evaluators of a large-scale, long-term school and educational-system management program implemented in Brazil’s public schools. There were three main challenges: (i) differential reactions to the evaluation from treated and control units, (ii) political cycles and bureaucracy turnover, and (iii) usual statistical issues, such as contamination, small sample sizes, and attrition. The first was solved by adopting outcome measures, whose performance incentives were independent of the treatment assignment. Dissemination of initial evaluation results among managers and school principals, who recognized the program’s benefits and demanded its continuation, was a way to deal with the second challenge. Finally, the long-term trust relationship built with implementers, school principals, and managerial staff of the state bureaucracy allowed evaluators to avoid contamination effects and abrupt program terminations, which would have reduced sample sizes. Although other countries deal with different contexts and specificities, the Brazilian case can anticipate recurring problems and offer insights into evaluating large-scale and high-complexity programs with similar characteristics.