Minimum Distance Methods for Models with Cointegration in Time Series and Short Panels

9720675埃利奥特这个项目的目标是开发用于协整时间序列回归和短面板数据模型的推断的最小距离方法。协整模型特别受到应用研究人员的青睐，因为它们与经济理论直接相关，即区分长期和短期动态。最小距离法似乎为估计和对这些模型进行推断提供了理想的方法。这是非常有用的，因为目前估计协整模型的方法已被证明很难扩展到对应用计量经济学家有用的方向。方法往往是协整问题本身所特有的，或者会导致复杂的计算方法。在大多数情况下，当应用最小距离法时，闭合形式的解可用于估计。这些方法非常简单和容易理解，使得能够相当简单地将协整模型估计方法扩展到应该在应用工作中被证明有用的方向。时间序列模型的特殊扩展是当存在对协整向量(线性或非线性，方程内或方程间)的限制时的估计，平稳变量的存在，以及异方差的存在。在面板模型中，该项目将在可获得对许多个人的观察时，在短时间维度面板数据集中提供和评估用于推断的估计值和规则。这个模型的重点是尽可能多地融入个体之间的异质性。这些方法也将向与时间序列模型相同的方向扩展。?？

9720675 Elliott The objective of this project is to develop minimum distance methods for inference in cointegrated time series regressions and short panel data models. Cointegrated models have found particular favor among applied researchers because of their direct relationship with economic theory, i.e., distinguishing between long-run and short-run dynamics. Minimum distance methods appear to provide ideal methods for estimating and conducting inference on these models. This is very useful as current methods of estimation of cointegrating models have proven quite difficult to extend in directions that are useful to applied econometricians. Methods are often special to the problems of cointegration per se or result in complicated computational methods. In most situations closed form solutions are available for estimation when minimum distance methods are applied. These methods are very simple and well understood enabling fairly simple extension of methods for estimation of cointegrated models in directions that should prove useful in applied work. The particular extensions envisaged for time series models are estimation when there are restrictions on the cointegrating vectors (linear or nonlinear, within or across equation), the presence of stationary variables, and the presence of heteroskedasticity. In the panel models, the project will be to provide and evaluate estimators and rules for inference in short time dimension panel data sets when observations on many individuals are available. The focus of this model is to incorporate as much heterogeneity across individuals as possible. These methods will be extended also in the same direction as the time series models. ??