Working Papers

This article presents identification results for the marginal treatment effect (MTE) when there is sample selection. We show that the MTE is partially identified for individuals who are always observed regardless of treatment, and we derive sharp bounds on this parameter under four sets of assumptions. The first identification result combines the standard MTE assumptions without any restrictions to the sample selection mechanism. The second result imposes monotonicity of the sample selection variable with respect to the treatment, considerably shrinking the identified set. Third, we incorporate a stochastic dominance assumption which tightens the lower bound for the MTE. Finally, we provide a set of conditions that allows point identification for completeness. Our analysis extends to discrete instruments and distributional MTE. All the results rely on a mixture reformulation of the problem where the mixture weights are identified. We therefore extend the Lee (2009) trimming procedure to the MTE context. We propose some preliminary estimators for the bounds derived, provide a numerical example and simulations that corroborate the bounds feasibility and usefulness as an empirical tool. In future drafts, we plan to highlight the practical relevance of the results by analyzing the impacts of managed health care options on health outcomes and expenditures, following Deb, Munkin, and Trivedi (2006).

Presented at the 2019 Bristol Econometrics Study Group.

2. Private Education Market, Information on Test Scores and Tuition Practices (with Rômullo Carvalho, Sergio Firpo and Vladimir Ponczek - Working Paper)

In this paper, we investigate the impact of information disclosure on price-quality relationship in the private school market in Brazil. We use the disclosure policy on scores of an exit exam in Brazil (ENEM) that took place in 2006. We construct a novel longitudinal data set on private schools that includes information on ENEM average scores in the years before and after its publication and on tuition fees for all years. We show that the correlation between test scores and price significantly becomes positive after the publication and increases over the years. Furthermore, this correlation becomes stronger for schools whose quality was noisily perceived previously.

Presented at the 36th Meeting of the Brazilian Econometric Society (2014), the VII Encontro CAEN-EPGE de Políticas Públicas e Crescimento Econômico (2015), the 20th Annual Meeting of the Latin American and Caribbean Economic Association (2015).

Published Papers

The synthetic control (SC) method has been recently proposed as an alternative method to estimate treatment effects in comparative case studies. Abadie et al. (2010) and Abadie et al. (2015) argue that one of the advantages of the SC method is that it imposes a data-driven process to select the comparison units, providing more transparency and less discretionary power to the researcher. However, an  important limitation of the SC method is that it does not provide clear guidance on the choice of predictor variables used to estimate the SC weights. We show that such lack of specific guidances provides significant opportunities for the researcher to search for specifications with statistically significant results, undermining one of the main advantages of the method. Considering six alternative specifications commonly used in SC applications, we calculate in Monte Carlo simulations  the probability of finding a statistically significant result at 5% in at least one specification.  We find that this probability can be as high as 13% (23% for a 10% significance test) when there are 12 pre-intervention periods and decay slowly with the number of pre-intervention periods. With 230 pre-intervention periods, this probability is still around 10% (18% for a 10% significance test). We show that the specification that uses the average pre-treatment outcome values to estimate the weights performed particularly bad in our simulations. However, the specification-searching problem remains relevant  even when we do not consider this specification. We also show that this specification-searching problem is relevant in simulations with real datasets looking at placebo interventions in the Current Population Survey (CPS).  In order to mitigate this problem, we propose a criterion to select among SC different specifications based on the prediction error of each specifications in placebo estimations.

Presented at the 2016 Africa Meeting of the Econometric Society, 38th Meeting of the Brazilian Econometric Society (2016) and the 2018 Young Economists Symposium.

2. Synthetic Control Method: Inference, Sensitivity Analysis and Confidence Sets (with Sergio Firpo - Published at the Journal of Causal Inference)

Recently, the Synthetic Control Estimator was proposed to answer questions involving counterfactuals when only one treated unit and a few control units are observed. Although this method was applied in many empirical works, the formal theory behind its inference procedure is still an open question. In order to fulfill this lacuna, we make clear the sufficient hypotheses that guarantee the adequacy of Fisher's Exact Hypothesis Testing Procedure for panel data, allowing us to test any sharp null hypothesis and, consequently, to propose a new way to estimate Confidence Sets for the Synthetic Control Estimator by inverting a test statistic, the first confidence set when we have access only to finite sample, aggregate level data whose cross-sectional dimension may be larger than its time dimension. Moreover, we analyze the size and the power of the proposed test with a Monte Carlo experiment and find that test statistics that use the synthetic control method outperforms test statistics commonly used in the evaluation literature. We also extend our framework for the cases when we observe more than one outcome of interest (simultaneous hypothesis testing) or more than one treated unit (pooled intervention effect) and when heteroskedasticity is present.

Presented at the California Econometrics Conference (2015), the 37th Meeting of the Brazilian Econometric Society (2015), the 2016 Latin American Workshop in Econometrics, and the 2017 Winter Meeting of the Econometric Society.

Won the Prize of Best Econometric Article presented at the 37th Meeting of the Brazilian Econometric Society (2015).

Code: We made available an R function and a Stata program to compute the confidence sets proposed in this paper. They can be downloaded here.

I apply the synthetic control method for Brazilian city-level data during the 20th Century in order to evaluate the economic impact of the Free Trade Zone of Manaus (FTZM). I find that this enterprise zone had positive significant effects on real GDP per capita and Services Total Production per capita, but it also had negative significant effects on Agriculture Total Production per capita. My results suggest that this subsidy policy achieved its goal of promoting regional economic growth at the cost of provoking mis-allocation of resources among economic sectors. They also reject the view that an industrialization policy will benefit all economic sectors due to positive spill-overs of the manufacture sector that are strong enough to compensate for the negative effect of the mis-allocation of resources.

O presente estudo objetiva analisar as características da demanda por planos de saúde no Brasil, averiguando os diversos atributos associados aos planos e à percepção dos consumidores sobre a qualidade. Para tal, foi implantada uma análise econométrica para as amostras da PNAD de 1998, 2003 e 2008. Essa base de dados nos permite analisar a evolução temporal da demanda por planos de saúde privados e como se relacionam a percepção dos indivíduos sobre as características do plano de acordo e o valor mensal pago. Os resultados apontam que o aumento no número de brasileiros segurados por planos de saúde privados se deve principalmente à evolução dos indicadores socioeconômicos das famílias. Por fim, há uma correlação positiva entre satisfação e preço dos planos, muito embora, para a maior parte dos atributos analisados, não pareça haver diferença sistemática entre os planos que justifique, por si só, essa correlação. Tal resultado aponta para a existência de possíveis diferenças entre os planos em atributos não mensurados nas bases de dados.