A0703
Title: Combined estimation of semiparametric panel data models
Authors: Bai Huang - Central University of Finance and Economics (China) [presenting]
Tae-Hwy Lee - University of California Riverside (United States)
Aman Ullah - University of California Riverside (United States)
Abstract: The properties of the combined (model averaging) estimation of semiparametric panel data models with endogeneity are examined. We examine the semiparametric (SP) panel data model with random effect (RE) and fixed effect (FE) and consider a combined estimator of SP RE and SP FE estimators. When the SP RE estimator suffers from inconsistency due to the random individual effect being correlated with the regressors. We show that under certain conditions, the SP combined estimator has strictly smaller risk than SP FE estimator. The asymptotic distribution and risk of the combined estimator are derived using a local asymptotic framework. The Monte Carlo study shows that the SP combined estimator outperforms better than SP FE and SP RE estimators except when the degree of endogeneity or heterogeneity is very small. An empirical application is also presented. According to our calculation of the asymptotic risks of the alternative estimators under comparison, the combined estimation allows researchers to implement efficient estimation under the presence of endogeneity without having to select one of efficient or consistent estimators. Even when there is no endogeneity or when endogeneity is strong, the selection of an efficient estimator or a consistent estimator can be conducted by the combined estimator, as the weights will then be 1 or 0. Hence, the combined estimator is an omnibus estimator across all degrees of endogeneity, especially useful when the endogeneity is weak.