9730277 Andrews This project involves research in three different areas of econometrics: (1) Extremum Estimators in Non-Standard Scenarios. A standard assumption for estimators and test statistics is that the true parameter is in the interior of the parameter space. This assumption is convenient because it allows one to make use of the fact that first order conditions hold, at least asymptotically. There are numerous cases of interest, however, in which the true parameter is on the boundary of the parameter space. This project develops methods for testing, model selection, bootstrap and subsampling procedures and Bayesian asymptotics for problems where this standard assumption no longer holds. (2) Moment and Model Selection for the Generalized Method of Moments (GMM). Empirical researchers using GMM often find that not all moment conditions are correct. At the same time, it is often the case that researchers have some uncertainty regarding the precise specification of the model of interest. For example, they may not know how many lags of a variable to include in the model or whether a variable should be included in the regressor or not. This project develops selection procedures for GMM estimators that simultaneously select correct moments and correct model specifications. These model/moment selection procedures are applied to dynamic panel data models with unobserved individual effects, an important area of applied econometrics. (3) Accelerated Bias-Corrected Confidence Intervals. Bootstrap methods have gained a great deal of popularity in empirical research. Although the methods are easy to apply, determining the number of bootstrap repetitions to employ is a common problem in the existing literature. Typically, this number is determined in a somewhat ad hoc manner. This is problematic, because one can obtain a different answer from the same data merely by using different simulation draws if the number of bootstrap repetitions is too small. This project develops a method of determining the number of bootstrap repetitions for accelerated bias-corrected confidence intervals. ??
