154th Seminar

Topic:Panel data, fixed effects, and the jackknife  

 

 

Presenter:    Geert Dhaene

 

 

Time:March 25, 2017(Saturday)10001130

 

 

Venue:  Room 205, Jiageng Building 2

 

 

Abstract:It is well known that maximum likelihood estimation of nonlinear panel models with fixed effects produce estimates that suffer from large biases and confidence intervals with poor coverage. The jackknife offers a simple method to reduce this bias and to obtain confidence intervals that are correctly centered under rectangular-array asymptotics. By splitting the panel into blocks, the method is explicitly designed to handle dynamics in the data, and yields estimators that  are straightforward to implement and can be readily applied to a range of models  and estimands. We provide distribution theory for estimators of model parameters  and average effects, present validity tests for the jackknife, and consider  extensions to higher-order bias correction and to two-step estimation problems.  (Based on joint work with Koen Jochmans, Sciences Po)