New Approaches for Dynamic Panel Data Analysis

该项目旨在开发具有因子误差或近似因子误差结构的面板数据模型的建模，估计和推断的新方法。 这种错误结构植根于经济学和金融学，通常的加性个体效应和时间效应是这种结构的特例。解释变量与未观测因子和因子负荷相关。忽略这种相关性会导致不一致的估计。PI将开发一种广义Mundlak投影，作为传统Mundlak-Chamberlin投影的替代方案，用于处理这种相关性。 广义预测是对传统方法的改进，其结果可能是如何处理解释变量和未观察到的效果之间的相关性的改进，这项研究的更广泛的影响主要来自于这些方法在更多应用环境中的潜在使用;面板数据被广泛用于政策评估。

The project seeks to develop new methods for modeling, estimation, and inference in panel data models with a factor error or approximate factor error structure. This error structure has roots in economics and finance, and the usual additive individual effects and time effects are special cases of this structure. Explanatory variables are correlated with unobserved factors and factor loadings. Ignoring this correlation leads to inconsistent estimates. The PI will develop a generalized Mundlak projection that is an alternative to the traditional Mundlak-Chamberlin projection for dealing with this correlation. The generalized projection is an improvement on the traditional method, and the result may be improvements in how to handle correlations between explanatory variables and unobserved effects.The broader impacts of this research are largely from the potential use of these methods in more applied settings; panel data are widely used for policy evaluation.