A Shrinkage Approach to Fixed Effects Estimation of Nonlinear Panel Data Models

• Activity Number: 411
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Dalia A. Ghanem*+
• Companies: University of California, San Diego
• Keywords: incidental parameters ; maximum likelihood ; bias correction ; shrinkage ; panel data

Abstract

Nonlinear panel data models with fixed effects are an important case in econometrics where the incidental parameter problem arises and the maximum likelihood estimator (MLE) is asymptotically biased. Bias correction of the MLE, as in Hahn and Newey (2004), achieves consistency without increasing the asymptotic variance. Asymptotically, this approach is equivalent to a shift of the MLE distribution by a constant equal to the asymptotic bias. The paper provides Monte Carlo evidence that such an asymptotic-shift approach coupled with estimation uncertainty often leads to situations where the realization of the bias-corrected estimator (BCE) is more biased than the MLE. A shrinkage estimator is proposed to combine the MLE and BCE in Hahn and Newey (2004). Monte Carlo results show that the shrinkage estimator leads to further bias reduction and increased accuracy of coverage probabilities.