Econometrics & Applied Economics Seminar - Bin Peng (Deakin University)

Title: Panel Data Models with Potentially Misspecified Unknown Factors

Abstract: Panel data models with interactive fixed effects have been well investigated and applied since Pesaran (2006) and Bai (2009). However, not much effort has been made to generalise the unobservable factor structure and relax the corresponding assumptions. In this study, we investigate the consequences of misspecifying the property of unknown factors of a parametric panel data model. We show that the interactive fixed effects estimator still achieves the global minimum even when the properties of unknown factors are misspecified. From the point of view of principle component analysis, we formalize some advantages/ disadvantages when regressors and factors possess different trending behaviours across time. Some rates of convergence and an asymptotic normality are established accordingly. In addition, we find that nonstationarity of the factors can help reduce the requirement of the sample size along the time dimension. Moreover, the investigation on misspecification extends the discussions of Section 4.2 of Bai et al. (2009). We verify our findings through extensive simulation studies, and investigate income elasticity of health care expenditure using data of OECD counties.