Date of Award
Spring 2024
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Economics
First Advisor
Haile, Philip
Abstract
This thesis comprises two chapters in industrial organization, focusing on econometric methods for estimating discrete choice demand models. The first chapter discusses how we can adapt, in a computationally feasible way, a weak-identification robust inference method that is otherwise computationally very expensive. The second chapter demonstrates how the conventional method of gauging the accuracy of the demand estimator may be incorrect when a well-known type of instrumental variable is employed, and studies how we can apply correlation-robust variance estimators as a remedy. The first chapter proposes an approach to inference on BLP-style demand models when instruments are potentially weak. I show how in practice one can adapt the two-step identification-robust procedure proposed by Andrews (2018) to such models. Direct application of this approach introduces substantial computational complexity, since it requires grid search over a potentially large parameter space. I provide conditions under which the time complexity of the procedure is reduced from the total number of parameters to the number of so-called "nonlinear parameters." Monte Carlo simulations reveal that the two-step confidence set, equipped with my dimension reduction technique, achieves the correct coverage probability. It is also shown that, although the technique is developed under the assumption of homoscedasticity, the resulting confidence set still performs well when the true structural errors are heteroscedastic. The second chapter shows how the common practice of calculating the standard errors of demand estimators can be incorrect when using Hausman instruments. The typical method ignores a correlation pattern arising from the essential endogeneity of the price, usually underestimating the true variance of the estimators as a consequence. I explore methods to robustly calculate the standard errors for some popular subclasses of Hausman instruments, including region-based and adjacency-based instruments. Monte Carlo simulations are conducted to evaluate their performance. The results suggest using the robust standard errors whenever the requirements for the correction are met, such as a large number of regions. It is also recommended to avoid using "national-level" Hausman instruments.
Recommended Citation
Lee, Jaewon, "Essays on Demand Estimation" (2024). Yale Graduate School of Arts and Sciences Dissertations. 1485.
https://elischolar.library.yale.edu/gsas_dissertations/1485
