Why Special Economic Zones? Using Trade Policy to Discriminate Across Importers (main appendix) (supplementary appendices)
Revise and resubmit (2nd round), American Economic Review

Tariffs are generally assumed to depend on the product being imported, not the identity of the importer. However, special economic zones (SEZs) are a common and economically important policy tool used around the world to lower tariffs on selected intermediate goods for selected manufacturers. In effect, this creates a two-tiered tariff system: a firm in an SEZ faces a lower rate on a given product than the prevailing tariff faced by most buyers. I show theoretically that discrimination across buyers of the same good is optimal from the policymaker's perspective whenever a tax is intended to raise prices for sellers. This applies even when buyers are homogeneous, and is distinct from standard price discrimination. I build a model that applies this insight to protectionist tariffs on intermediate goods, and demonstrate that this mechanism motivates the use of SEZs. The model has substantial empirical traction, making specific predictions about about the form, composition, and size of zones in equilibrium. I investigate these predictions in a comprehensive new dataset on U.S. SEZs that I compiled from public records, and find that the implementation of zone policy is consistent with the theory, including in the way that it develops over time in response to exogenous changes in the import supply of particular intermediate goods.

Where Do the Data Come From? Endogenous Classification in Administrative Data [draft available on request]

Many types of data are organized according to codes that aggregate observations that share a set of observable characteristics. It is well known that aggregated data can create estimation problems; this paper focuses on the determinants of the aggregation, and shows that the degree of aggregation (and therefore bias) may be systematically related to characteristics of interest to both the econometrician and the policymaker. The level of aggregation can be understood as a choice of the classifier who uses the classification system to target policies and collect information, and weighs the benefits of better targeting against the higher costs of establishing and enforcing a finer classification. Consequently, we should expect the level of aggregation to vary across the classification schedule, and the extent of estimation problems related to aggregation to be correlated with the objective of the classifier. I present empirical evidence that this is a problem in practice, using information from the classifiation of traded goods in the U.S. In particular, I show that bias arising from aggregation in the estimation of elasticities calculated following Feenstra (1994) is negatively correlated with the level of tariff and the level of the elasticity. I also show this bias can change the correlation between tariffs and elasticities in qualitatively important ways. Finally, I present a technique in which, if the form of the classifier's objective can be inferred, then classification costs and the marginal benefit of classification can be jointly estimated using the observed response of the classification system to shocks. This in turn permits the degree of aggregation across the classification to be estimated and estimates of parameters of interest using the classification to be corrected.

Work in Progress:

Cutting Out the Middleman: The Structure of Chains of Intermediation (with Meredith Startz) (draft available soon!)

Foxes and Hedgehogs: The Determinants of Labor Market Adjustment Costs (with Hannes Schwandt and Sharon Traiberman)

Special Economic Zones in Developing Countries