Bayesian Inference and Econometric Modeling

该项目继续开发和应用贝叶斯统计方法来解决重要的经济问题。这位研究人员过去的工作产生了许多强大的分析工具，现在这些工具被广泛应用于经济学理论和实证研究的几乎所有领域。目前的项目侧重于寻找明确的、可重复的程序来产生观测模型，信息处理规则来结合模型和其他信息，以及使用数据来评估替代模型的程序。这项新研究的一个贡献来自于将当前构建经济模型的实践正式化，并以一种可重现和科学严谨的方式对其进行评估。使用这些方法的初步工作表明，贝叶斯方法可以从美国季度宏观经济数据和欧洲数据提供显著改善的宏观经济预测。提高经济预测的科学严谨性和经验准确性是全球变化经济学倡议的一个目标。例如，该项目正在开发的方法被用来组合和评估与气候变化的四个主要模型截然不同的预测。该项目的主要目标是为计量经济推断和建模问题提供一个统一的Bayes-Maxent方法，并检验其在宏观计量经济建模和预测以及生产函数分析两个应用领域的表现。在Bayes-Maxent方法中，观测模型及其相关的先验密度被导出为使用熵概念的显式的、受约束的最优化问题的解。此外，结合先验密度和似然函数的信息处理规则，例如贝叶斯定理，也被导出为约束极大值问题的解。进一步研究了Bayes-Maxent方法的特征，并扩展了其适用性。

This project continues to develop and apply Bayesian statistical methods to important economic problems. Past work by the investigator yielded many powerful analytical tools that are now widely used in almost every area of theoretical and empirical research in economics. The current project emphasizes finding explicit, reproducible procedures for producing models for observations, information-processing rules for combining models and other information, and procedures for evaluating alternative models using data. One contribution of this new research comes from formalizing current practice in constructing economic models and evaluating them in a way that is reproducible and scientifically rigorous. Preliminary work with these methods shows that the Bayesian approach can provide dramatically improved macroeconomic forecasts from U.S. quarterly macroeconomic data and from European data. Improving the scientific rigor and the empirical accuracy of economic forecasts is a goal of the economics of global change initiative. For example, the methods being developed by this project are being used to combine and evaluate the very different forecasts from the four primary models of climate change. The main objective of this project is to provide a unified Bayes- Maxent approach to econometric inference and modeling problems and examine its performance in two areas of application, macro- econometric modeling and forecasting and production function analysis. In the Bayes-Maxent approach, models for observations and their associated prior densities are derived as solutions to explicit, constrained, optimization problems using entropy concepts. Further, information-processing rules which combine prior densities and likelihood functions, for example Bayes's Theorem, are also derived as solutions to constrained maxent problems. Research to characterize further and extend the applicability of the Bayes-Maxent approach are pursued.