Essays on International Economics and Applied Econometrics

Abstract

This dissertation comprises three essays on the economics of international trade and econometrics of treatment effects. Chapter One, based on work with Xavier Jaravel, studies the distributional effects of international trade in the United States. International trade may affect different groups of Americans differently because they have different exposure to it: some consume more imports and would benefit more if import prices go down, or they work in exporting industries, while others have their jobs threatened by import competition. We build detailed data to measure who is exposed more. We also develop a tractable model to translate these patterns into the counterfactual effects of trade liberalizations through prices and wages. We find that, contrary to prior research, the benefits of trade through falling prices are distributed equally between college and non-college educated consumers and across income groups. At the same time, trade favors college graduates through wages, so the net gains from trade liberalization are 16% higher for them. The other two chapters develop methodological tools for applied economists. Chapter Two, also based on work with Xavier Jaravel, considers event studies—commonly used research designs where all units in a panel receive treatment (for example, all states adopt some law), but the timing is random. Event studies are often viewed as analogous to difference-in-differences designs. This chapter shows that this analogy is misguided in two ways. First, standard tests for pre-trends are not feasible because of fundamental underidentification: all linear pre-trends are observationally equivalent in the data. The chapter shows how to test for non-linear pre-trends and how the underidentification problem can be resolved by imposing different restrictions on the model. Second, a simple regression of the outcome variable on the treatment dummy and unit and time fixed effects, that is ubiquitously used to summarize treatment effects, does not produce a desired result. This regression weights long-term effects negatively, which can result in a substantial bias. We propose estimators which do not suffer from this problem. The usefulness of the approach is demonstrated with an application to estimating the marginal propensity of consume out of tax rebates. Chapter Three moves the focus from how to estimate treatment effects to what to estimate. While most treatment evaluations report average effects of the treatment on a certain outcome, this chapter shows how to analyze the extensive margin—the population share of subjects (“responders”), for whom the outcome depends on whether they have been treated. This share cannot be point-identified because potential outcomes with and without treatment are never observed for the same subject. However, I characterize the sharp lower bound on this share. In the simplest randomized control trial case, it equals the total variation distance between the distributions of outcomes in the treatment and control groups. I show that this bound is informative using three applications related to behavioral biases in retirement savings choices, election fraud, and cheating in high-school standardized tests.