A0191
Title: Two-step instrumental variable estimation of linear panel data models with interactive effects
Authors: Guowei Cui - Huazhong University of Science and Technology (China)
Milda Norkute - Lund University (Sweden)
Vasilis Sarafidis - Monash University (Australia)
Takashi Yamagata - University of York (United Kingdom) [presenting]
Abstract: Instrumental variable (IV) estimators are proposed for linear panel data models with interactive effects in the error term and regressors. The IVs are transformed regressors, and it is not necessary to search for external IVs. We consider the models with homogeneous slopes and heterogeneous slopes. The approach asymptotically eliminates interactive effects in the error term and the regressors \textit{separately}. Asymptotic properties of the proposed estimators are investigated, which reveal that: (i) the $\sqrt{NT}$-consistent second-step estimator is free from asymptotic bias, which could arise due to the correlation between the regressors and estimation errors of interactive effects; (ii) under the same condition other existing estimators, which asymptotically eliminate interactive effects in the error term \textit{only} or in the regressors and error term \textit{jointly}, can suffer from asymptotic biases, and; (iii) the estimator is asymptotically as efficient as the latter estimator after the bias-correction, but the relative efficiency to the former estimator after the bias-correction is indeterminate.
