Model Averaging PartiaL Effect (MAPLE) Estimation with

Large Dimensional Data

报告人：Yundong Tu（University of California）

时 间：10-11：30am,Feb.13^th(Mon)

地 点：2号楼217

Abstract

This paper studies the estimation of the marginal effect of one economic variable on another in the presence of large amount of other economic variables| a problem frequently faced by applied researchers. The estimation is motivated via model uncertainty so that random components should be included to describe the economy according to the state of the world. A condition named \Conditional Mean Independence" is shown to be sufficient to identify the partial effect parameter of interest. In the case that the parameter of interest can be identified in more than one approximating model, we propose two estimators for such a parameter: generalized-method-of-moment-based model averaging partial effect (gMAPLE) estimator and entropy-based model averaging partial effect (eMAPLE) estimator. Consistency and asymptotic normality of the MAPLE estimators are established under a suitable set of conditions. Thorough simulation studies reveal that MAPLE estimators outperform factor based, variable selection based and other model averaging estimators available in the literature. An economic example is taken to illustrate the use of MAPLE estimator to evaluate the effect of inherited control on _rms' performance.Key Words: Partial Effect; Treatment Effect; Model Averaging; Bayesian Model Averaging; Jackknife Model Averaging; FOGLeSs; Variable Selection; Factor Models; InheritedControl.