Heteroskedastic supply and demand estimation: Analysis and testing

The Feenstra (1994) method is widely used in the international trade literature to estimate supply and demand elasticities. The method is mechanically an IV strategy, and we demonstrate that this has important implications for its application and reliability. The assumptions needed for it to yield unbiased estimates are stronger than previously understood, and in practice, estimates are subject to bias due to both weak instruments and violations of the exclusion restriction. We illustrate how these arise in context and show that standard tests identify estimates that are likely to be biased. In an application to U.S. import data, estimates of import demand and export supply elasticities are substantially lower among goods that pass standard tests relative to those that fail. We find evidence that this difference in elasticities reflects reduction of bias as well as some selection in the set of goods that tend to pass both tests.