My dissertation studies semi- and non-parametric estimation strategies for the distribution of heterogeneous causal effects with applications to labor economics and macroeconomics. \par

In the first chapter, I propose a nonparametric strategy to identify the distribution of heterogeneous causal effects. A set of identifying restrictions proposed in this chapter differs from existing approaches in three ways. First, it extends the random coefficient model by allowing potentially non-linear interaction between distributional parameters and the set of covariates. Second, the treatment effect distribution identified in this chapter offers an alternative interpretation to that of the the rank invariance assumption. Third, the identified distribution lies within a sharp bound of distributions of the treatment effect. An estimator exploiting the identifying restriction is developed by extending the classical version of statistical deconvolution method to the Rubin causal framework. I show that the estimator is uniformly consistent for the distribution of causal effects. \par

In chapter two, I apply the nonparametric method developed in the previous chapter to the estimation of heterogeneous effects of displacement on earnings losses. Using the Current Population Survey (CPS) individual-level data from 1996 to 2016, I show that the decline in labor incomes of displaced workers is not only substantial in magnitude compared to their non-displaced counterparts, but also varies significantly within groups characterized by, for example, tenure and educational attainment. I find that displaced workers, on average, lose 19% of their potential earnings while the dispersion of losses among workers is wide. In addition, estimated quantile effects of displacement are more dispersed when the local unemployment rate is higher. \par

In the third chapter, co-authored with Guido Kuersteiner, we develop a new asymptotic theory for flexible semi-parametric estimators of dynamic causal effects in data with discrete policy interventions. Our framework extends existing theory of propensity score weighted estimators to weakly dependent processes. We show uniform consistency and asymptotic normality by applying a newly-developed asymptotic theory for the series estimator over a non-compact support. The estimator proposed in this chapter captures non-linear and asymmetric impulse response functions that are often difficult to be accommodated in parametric models.