Inference on Extremes in Economic Regression Analysis

经济回归分析中的极值推断

• 批准号: 0649388
• 负责人: Victor Chernozhukov
• 金额: $ 7.21万
• 依托单位: Massachusetts Institute of Technology
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2007
• 资助国家: 美国
• 起止时间: 2007-02-01 至 2008-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0649388&HistoricalAwards=false
• 关键词: Inference; Extremes; Economic; Regression; Analysis

The purpose of the project is to change dramatically the methods and software available for studying extreme phenomena in economic regression analysis. More specifically, the project develops feasible and practical inference methods for extremal quantile regressions - the models that describe how regressors X affect the outcome of interest Y in the tails. For example, extremal quantile regressions describe how mothers' socio-economic characteristics (smoking, health-care, education) affect the very low quantiles of birthweights. The project makes it possible to make inferences about such regression relationships. The project develops inference methods for parametric and non-parametric extremal quantile regression models and public software (based on the freeware environment) that implements these methods. The project is strictly focused on the following three parts:(1) Inference methods in linear extremal quantile regression,(2) Inference methods in nonparametric extremal quantile regression,(3) Software package (free and publicly available) which implements (1) and (2). The software will be available as a documented R-package with help files and a vignette. The vignette is the document that will contain a short description of the methods, followed by tutorial examples using interesting data sets.Broader Impact of the Proposal: As stated above, one of the purposes of the project is to produce public software that implements inference methods for extremal quantile regression. After the software package is available, any practitioner will be able to download and install the package from R-website www.r-project.org and use it in the analysis of any data-set, aided by help files and tutorial examples. It should be mentioned that the reason for choosing R as a programming environment is that it serves as a common, free statistical environment widely used by researchers (econometricians, statisticians, biometricians) and practitioners. This characterizes the broader impact of the proposed project. The project will also have direct educational impact by involving help of two students: one undergraduate student and one graduate student. The graduate student will be a coauthor, and both students will be co-authors and co-developers of the software package.

该项目的目的是极大地改变可用于研究经济回归分析中极端现象的方法和软件。更具体地说，该项目为极端分位数回归开发了可行和实用的推理方法-描述回归变量X如何影响尾部感兴趣的Y结果的模型。例如，极端分位数回归描述了母亲的社会经济特征(吸烟、医疗保健、教育)如何影响出生体重的极低分位数。该项目使得对这种回归关系进行推断成为可能。该项目开发了参数和非参数极值分位数回归模型的推断方法，以及实现这些方法的公共软件(基于免费软件环境)。(1)线性极值分位数回归的推断方法，(2)非参数极值分位数回归的推断方法，(3)实现(1)和(2)的软件包(免费和公开)。该软件将作为一个有文档记录的R包提供，其中包含帮助文件和小插曲。该文件将包含方法的简短描述，然后是使用有趣的数据集的教程示例。提案的广泛影响：如上所述，该项目的目的之一是生成实现极值分位数回归的推理方法的公共软件。在该软件包可用后，任何从业者都可以从R-WebSite www.r-project t.org下载和安装该软件包，并在帮助文件和教程示例的帮助下将其用于任何数据集的分析。值得一提的是，之所以选择R作为编程环境，是因为它是研究人员(计量经济学家、统计学家、生物计量学家)和从业人员广泛使用的通用、免费的统计环境。这是拟议项目的更广泛影响的特征。该项目还将通过两名学生的帮助产生直接的教育影响：一名本科生和一名研究生。研究生将是该软件包的共同作者，两名学生将是该软件包的共同作者和共同开发人员。