Flexible Bayesian Econometric Modeling

拟议研究的目标是在两个方向上推进计量经济模型。在第一个方向，开发的工具(A)显著扩大了可用于测试和估计经济行为模型的数据集的范围，(B)大大减少了在新制定的模型中进行正式统计推断所需的时间和精力。TN.第二个方向是制定计量经济学模型，这些模型(A)施加更接近经济理论的假设，以及(B)允许在经济理论不涉及的那些模型方面比现在有更大的灵活性。第一个方面是，计量经济学模型中的推断是贝叶斯的，以数据为条件，并保持与似然原则的一致性。第二个方面是使用模拟方法来解决无法分析攻击的关键技术问题。在第一个前进方向中，将使用来自先验和数据密度的正向模拟，以允许根据数据的选定特征对感兴趣的模型、参数和函数进行推断。使用数据的选定特征的能力允许在可使用的数据集的宇宙中进行扩展。与马尔可夫链蒙特卡罗或模拟矩方法等反向模拟方法不同，这些前向模拟器非常容易构造。这种简单性大大减少了计量经济学建模的开发时间。在第二个前进方向，将使用非常灵活的分布假设来允许数据驱动推理，而不是任意但技术上方便的假设。这些假设包括条件依赖于协变量或状态变量的多变量正态混合模型。经济理论的限制，如凹性，可以通过适当的先验分布和马尔可夫链蒙特卡罗后验模拟方法施加在这些模型中。该项目的目标显著增强了经济学以及社会学和政治学等相关学科的实证工作能力。该项目的成果将包括免费提供的最先进的、用户友好的软件。该软件将成为该项目和相关项目正在开发的基于网络的教育和软件系统的一部分。研究生对项目的参与对这些成果至关重要，并将对这些学生的论文研究起到补充作用。

The objective of the proposed research is to advance econometric modeling in two directions. In the first direction, tools are developed that (a) significantly expand the universe of data sets that can be used for testing and estimation models of economic behavior and (b) drastically reduce the time and effort required to conduct formal statistical inference in newly formulated models. Tn. the second direction, econometric models are formulated that (a) impose assumptions that are closer to those of economic theory and (b) permit greater flexibility in those aspects of models to which economic theory does not speak, than is now the case.Two aspects of the methods to be employed are common to these objectives. The first aspect is that inference in econometric models is Bayesian, conditioning on the data and maintaining consistency with the likelihood principle. The second aspect is that simulation methods are used to solve key technical problems that cannot be attacked analytically. In the first direction of advance, forward simulation from prior and data densities will be used to permit inference about models, parameters, and functions of interest conditional on selected features of the data. The ability to use selected features of the data enables expansion in the universe of data sets that can be used. As opposed to backward simulation methods like Markov chain Monte Carlo or simulated method of moments, these forward simulators are very easy to construct. This simplicity greatly reduces the development time in econometric modeling. In the second direction of advance, very flexible distributional assumptions will be used to permit the data, as opposed to arbitrary but technically convenient assumptions, to drive inference. These assumptions include multivariate normal mixture models with conditional dependence on covariates or state variables. Restrictions from economic theory, such as concavity, can beimposed in these models by means of appropriate prior distributions and Markov chainMonte Carlo posterior simulation methods.The objectives of the project significantly enhance the capabilities for empirical work in economics, and related disciplines like sociology and political science. The outcome of the project will include state-of-the-art, user friendly software that will be made freely available. This software, in turn, will become part of a web-based educational and software system being developed in this and related projects. Graduate student participation in the project is essential to these outcomes, and will be complementary tothese students' dissertation research.