Econometric Methods for Structural and Semiparametric Models

点击翻译按钮获取中文摘要

ABSTRACTPROPOSAL NO: 0351259INSTITUTION: Harvard UniversityNSF PROGRAM: EconomicsPI: Porter, Jack RTITLE: Econometric Methods for Structural and Semiparametric ModelsThis research deals with econometric methods for various economic structural and semiparametric models. The research consists of several projects. The first project is an empirical project on multi-unit auctions in the deregulated electric utility market in England and Wales. Bid data and the actual auction software will be used to estimate fixed costs and costs of capital, which will be used to analyze incentives to invest, auction design issues, and possible collusive behavior. Companion work develops a new approach to dynamic auction model estimation using the first order condition derived from a bidder 's optimization problem as a moment for estimation and avoiding the computational difficulty associated with numerically solving for optimal bidding strategies. This research provides empirical results relevant to the recent debates concerning electricity markets, and, more generally, will provide new tools for the empirical analysis of many other auction markets.The second project considers certain semiparametric models that are typically estimated using an initial "plug-in "conditional expectation estimator and pursues an approach to estimation that avoids using a sample size dependent smoothing parameter. A nearest neighbor approach is combined with local polynomial regression to achieve sufficient bias-rate reduction for root n consistency even with a high dimensional covariate space. Leading applicable models include the partially linear model and the average treatment effect framework. This research is aimed at providing econometric tools that are easy for the empirical researcher to apply within a class of semiparametric models. The third Project uses Le Cam 's limits of statistical experiments theory to generalize standard results on efficient point estimation in parametric models. It shows how to construct a local shift to the MLE that yields efficiency under more general conditions, including all regular and many leading nonregular models along with symmetric or asymmetric loss. This work provides best estimators in these models, which are the workhorse of empirical research in many fields. The third project extends semiparametric estimation of treatment effects in the regression discontinuity design to allow for an unknown cut-off in treatment assignment. This work also characterizes estimation in a nonparametric analog of the parametric unknown structural breakpoint model. This approach will further broaden the applicability of the regression discontinuity framework. These projects are aimed at providing new econometric tools for the applied researcher.