Research
Working Papers
Direct and Spillover Effects of Generative AI: Evidence from the Judiciary (JMP, link coming soon)
Generative AI lowers the cost of writing text that someone else must read and act on. This paper measures the impacts of this shock in airline litigation cases in the small-claims courts of São Paulo, Brazil, where it affects both ends of the production chain. A validated detector tracks the start of AI usage among lawyers and judges for cases filed between 2022 and 2025. Using within-lawyer difference-in-differences comparisons of adopters and non-adopters, I show that (i) adopting lawyers increase filings: among those active in 2022, they are 17 percentage points more likely to file in a given month. (ii) They file longer petitions and use more complex language, but their additional lawsuits are similar in merit and final outcome rates. (iii) Judges are impacted by lawyers’ AI adoption: cases filed by adopting lawyers are 5 to 8 percentage points less likely to reach a first sentence within six months. AI adoption among judges does not affect the dispersion of their judicial decisions, but there's suggestive evidence that it improves their productivity. Results are in line with cheaper writing increasing the workload of courts through more claims of similar merit instead of a flood of weaker cases, while increasing the time taken to reach a decision.
Payroll Tax, Employment and Labor Market Concentration (with Raphael Corbi and Renata Narita) Revise and resubmit, Labour Economics. IZA Discussion Paper 18838 · Winner of the Applied Microeconomics prize, Brazilian Econometric Society, 2024.
How much employment can be generated by decreasing payroll taxes? We examine this question by exploring the staggered rollout of a large payroll tax reform in Brazil. Using administrative matched employer-employee data, we find an increase of 5 percent on employment due to both firm growth and firm entry, no impact on wages and a significant increase in profits. Moreover, employment effects are driven by less concentrated labor markets, consistent with predictions from an oligopsony model.
The Electoral Effects of Conditional Cash Transfers
This paper examines a large quasi-exogenous expansion of Brazil’s conditional cash transfer program, Bolsa Família, which increased beneficiary enrollment by roughly 13% in affected municipalities. Results are only partially aligned with the usual connection between social spending and incumbent support: we show that the expansion boosted voter turnout but did not produce lasting gains in incumbent vote shares. Evidence of short-term electoral returns is observed only in the presidential election immediately following the benefit expansion, with no significant effects in subsequent executive elections at federal, state and local levels. We also find no evidence that political polarization explains the weak long-run response. To test the asymmetry of these effects, we then proceed to explore a 2013 national eligibility check that removed over 700,000 families from the program. These removals are correlated to higher rates of blank voting and reduced support for the incumbent party, suggesting that losses in social benefits may carry sharper electoral consequences than equivalent gains.
The Chosen, the Cool and the Central: Peer Education and Teen Pregnancy in Brazil (with Emily Breza, Eliana La Ferrara, Pedro Rosa Dias and Victor Orozco)
We conduct a field experiment in Bahia, Brazil to test whether a peer-led program for secondary school students can improve sexual health knowledge, increase contraceptive use and reduce teenage pregnancy. Because peer educators are key to the program’s success, we compare three strategies for selecting them. Students must engage with the program, directly or through word of mouth, and must find its messages credible and socially palatable. We therefore compare selection by (i) school administrators, as a policy-relevant benchmark; (ii) network centrality, to capture the mobilizers’ reach; and (iii) popularity, to capture social status and norm-setting ability. We find that the intervention significantly increases contraceptive use and reduces teenage pregnancy. Consistent with the program decreasing the riskiness of sexual behavior, treated students also become sexually active sooner than controls. Impacts depend crucially on the selection strategy: administrator-based selection performs substantially worse than centrality- and popularity-based selection. Using a model of diffusion and adoption with school-level externalities, we show that centrality is particularly effective in spreading information and knowledge, while popularity and centrality together shape responses in sexual activity and contraceptive use.
Political Brokers and the Allocation of Social Benefits: The Case of Bolsa Família (with Caio Pedro Castro, Thea How Choon and Giovanna Marcolongo)