CAREER: Bayesian Econometric Modeling and Nonparametric Identification

该项目旨在引入和改进分析经济数据和强有力地评估政策可能影响的方法，并开发新的课程材料来培训研究生分析经济数据。该项目的第一部分将开发新的经济数据灵活模型的估计方法，通过采用贝叶斯方法进行推理来处理估计问题，在估计问题中，先验知识仅对可能性的范围施加有限的限制--非参数估计问题和半参数估计问题。贝叶斯方法之所以有吸引力，是因为它们具有风险最优特性，而且它们可以用来生成包含参数不确定性的未来结果的预测分布。研究人员以前在纵向数据模型中的半参数贝叶斯推断方面的工作将被扩展，以允许额外的回归变量、二元结果和未知干扰密度的替代表示。许多政策问题可以被框架为关于根据潜在结果定义的治疗效果的问题。该项目的第二部分将应用灵活的贝叶斯方法来估计涉及治疗效果的问题，并将这种方法与更传统的方法进行比较，以了解这些治疗效果。这项研究的一部分将考虑连续治疗的治疗效果的推断，当治疗分配与潜在结果无关时，条件是预处理变量向量。这项研究的另一个方面是将早期关于因果工具变量估计的参数贝叶斯方法的研究扩展到半参数模型。本项目的第三部分侧重于将计量经济学方法论的最新发展带入课堂，以便更好地培训研究生使用实用、灵活的实证方法。

This project aims to introduce and improve methods for analyzing economic data and robustly assessing the likely impact of policies, and to develop new course material to train graduate students in the analysis of economic data. The main approach centers around Bayesian estimation methods applied to large parameter spaces, in combination with recent strategies for defining and identifying effects of treatments or policies.The first component of this project will develop new methodology for estimation of flexible models for economic data by adapting Bayesian methods for inference to handle estimation problems in which prior knowledge places only limited restrictions on the range of possibilities-nonparametric and seimparametric estimation problems. Bayesian methods are attractive because they have risk optimality properties, and because they can be used to generate predictive distributions for future outcomes that incorporate parameter uncertainty. Previous work by the investigator on semiparametric Bayesian inference in models for longitudinal data will be extended to allow for additional regressors, binary outcomes, and alternative representations of unknown disturbance densitiesMany policy questions can be framed as questions about treatment effects defined in terms of potential outcomes. The second component of this project will apply flexible Bayesian methods to estimation problems involving treatment effects, and compare this approach to more conventional methods for learning about these treatment effects. One part of' this research will consider inference for treatment effects with continuous treatments when treatment assignment is independent of potential outcomes conditional on a vector of pretreatment variables. Another aspect of the research is to extend earlier research on parametric Bayesian approaches to causal instrumental variables estimation to semiparametric models.The third component of this project focuses on bringing selected recent developments in econometric methodology into the classroom in order to better train graduate students in practical, flexible empirical methods.