Research

Instrumental Variables with Multiple Time Periods (Job Market Paper). 

Abstract: The instrumental variables (IV) method has been widely studied in cross-sectional settings. However, many practical applications involve panel data, in settings where a unit's treatment status may turn on or off over time. I show that in the presence of dynamic treatment effects, that is, if past treatments affect current potential outcomes, standard methods are no longer valid if the instruments are serially correlated. This paper proposes the nonparametric identification of dynamic causal effects in a potential outcomes framework in which potential outcomes depend on the treatment path taken by a unit through time but the first stage is static, in the sense that at each period the IV only instruments its contemporary treatment. I provide a nonparametric estimator that is consistent over the randomization distribution and derive its finite population limiting distribution as the sample size increase.  Monte Carlo Simulations illustrate the desirable finite-sample properties of the estimators. An application of the estimator show that law enforcement curbs illegal deforestation until two years after the actual detection of illegal practice, but effects fade out afterwards.

Marginal Treatment Effects in Difference-in-Differences  (with Cristine Pinto) . Submitted

You can find me talking briefly about the paper here!

Abstract: Difference-in-Differences (DiD) is a popular method used to evaluate the effect of a treatment that exploits variation in treatment status that comes from the exposure to a shock, usually in the form of a policy change. When there is imperfect compliance towards the shock, the usual DiD estimand fails to recover relevant causal parameters. This article presents an identification strategy in DiD settings with imperfect compliance that identifies Marginal treatment effects (MTE). We show how to combine and modify standard instrumental variables (IV) and DiD assumptions to identify treatment effects in DiD settings where individuals enter into treatment with at least partial knowledge of their unobservable gains. We propose two estimators for the MTE that are consistent under different assumptions regarding the functional form of potential outcomes and prove their asymptotic normality. Furthermore, we derive an estimator for the local average treatment effect (LATE) that is robust to misspecification of the MTE model. We assert the desirable finite-sample properties through simulation studies of a linear MTE model. Finally, we use our results to investigate heterogeneity in the returns to primary education attendance in Indonesia. We find a pattern of reserve selection gains in the returns of primary education. Individuals with a higher distaste for enrolling in primary education have positive returns, while individuals with a lower distaste may have a negative return to primary education.

Difference-in-Discontinuities: Estimation, Inference and Validity Tests (with Stephanie Shinoki and Cristine Pinto).

Abstract: This paper sets out to explore the econometric theory behind the newly developed difference-in-discontinuities design. Despite its increasing use in causal inference research, there are currently limited studies of its underlying principles and properties. The method combines elements of regression discontinuity and difference-in-differences, allowing researchers to eliminate the effects of potential confounders at the discontinuity threshold and account for changes in the larger environment. We formalize the difference-in-discontinuity theory by stating the identification assumptions under which the difference-in-discontinuities estimate is valid and propose two nonparametric estimators, deriving their asymptotic properties. Monte Carlo simulation studies show that the estimators have good performance in finite samples. Finally, we revisit a difference-in-discontinuities paper that studies the effects of relaxing fiscal rules on public financial outcomes in Italian municipalities. The results show that the proposed estimator exhibits substantially smaller confidence intervals for the estimated effects.

Publications