EcoSta 2019: Start Registration
View Submission - EcoSta2019
Title: Averaging estimators for heterogeneous dynamic panel regressions with weakly exogenous regressors and multifactor error Authors:  Chang-Ching Lin - National Cheng Kung University (Taiwan) [presenting]
Shou-Yung Yin - National Taipei University (Taiwan)
Abstract: Model averaging is considered in heterogeneous dynamic panel regressions with weak exogenous regressors and a multifactor error structure. Under a local to zero framework, we show that the common correlated effects mean group (CCEMG) estimator exists three different components of the asymptotic bias, the fundamental bias from ignoring variables, time series bias and the bias from the truncation of number of lags of augmented regressors used to approximate unobserved factor structure. We then propose a focused information criterion and a plug-in averaging estimator based on the half-panel jackknife bias-corrected CCEMG estimators for the full model and all submodels. Since the fundamental bias and the bias from the truncation cannot be corrected, the trade-off between bias and variance exists, and the proposed methods can minimize the asymptotic mean squared errors. Monte Carlo simulations show that the proposed averaging method generally outperforms other methods. An empirical study on the relationship between the commodity price volatility and the economic growth is considered as an application.