题目：A Flexible Test For Conditional Independence(Job Market Paper)

报告人：Meng Huang

Department of Economics

University of California, San Diego

时间：2009年2月17日上午8：00-9：30

地点：成人直播新楼217教室

Abstract

This paper proposes a .exible test for conditional independence, which is simple to implement yet powerful in the sense that it is consistent and achieves pn local power. The test statistic is based on an estimator of the topological .distance. between restricted and unrestricted probability measures corresponding to conditional independence or its absence. The distance is evaluated using a set of Generically Comprehensively Revealing (GCR) functions such as exponential or logistic functions, which are indexed by nuisance parameters.The use of GCR functions makes the test able to detect any deviation from the null. I use a kernel smoothing method when estimating the distance. I obtain a single test statistic by integrating out nuisance parameters. And I simulate the critical values by a conditional simulation approach. Monte Carlo experiments show the test performs well for .nite samples. Finally, I test the key assumption of unconfoundedness in the context of estimating the returns to schooling.

题目：Parametric Efficient Estimation in Threshold Regression",

Job Market Paper I, November 2007

报告人：Ping YU（UW Madison）

时间：2009年2月17日上午9：30-11：00

地点：成人直播新楼217教室

ABSTRACT: This paper considers efficient estimation in the framework of parametric discontinuous threshold regression with iid data. The main result is that the Bayesian estimator is asymptotically efficient among all estimators in the locally asymptotically minimax sense. In particular, the Bayesian estimator of the threshold point is asymptotically strictly more efficient than the left-endpoint maximum likelihood estimator and the popular least squares estimator. This phenomenon is due to the asymptotic nonsufficiency of the left-endpoint maximum likelihood estimator and is not surprising when the threshold point is treated as a "middle" boundary. There are also two additional contributions from this paper. The first of these is a new estimator called the middle-point maximum likelihood estimator for the threshold point, whose efficiency is between the left-endpoint maximum likelihood estimator and the Bayesian estimator in most cases. The second additional contribution is to provide new algorithms to calculate the asymptotic distribution and risk for the left-endpoint maximum likelihood estimator, middle-point maximum

likelihood estimator, posterior mean, and posterior quantile estimators of the threshold point.

Chapter 3

Job Market Paper II, October 2008

ABSTRACT: In this paper, I put forward an estimator called the semiparametric empirical Bayesian estimator for the threshold point in discontinuous threshold regression. This estimator is adaptive to the unknown distribution of the data even under standard conditional moment restrictions and is semiparametrically efficient in the locally asymptotically minimax sense. The intuition for this result is that the threshold point is essentially a "middle" boundary of the threshold variable, and that the information used to estimate a boundary and regular parameters are independent. This paper's result is very different from results in regular models. Furthermore, I use the nonparametric posterior interval to construct a confidence interval for the threshold point. Such construction could be treated as an alternative of the subsampling method and Hansen's method (2000). The simulations and application show that the semiparametric empirical Bayesian method and its recursive refinement are substantially better than the existing methods.