【Metrics Seminar】Partial Identification of Nonlinear Models with Misclassification Error: A Perturbation Approach
2024/12/05
研討會日期 : 2024-12-05
時間 : 14:30
主講人 : Professor Ji-Liang Shiu (徐吉良)
地點 : Conference Room C103
主持人 : Professor Yu-Chin Hsu
演講者簡介 : Professor Ji-Liang Shiu received his Ph.D. in Economics from Johns Hopkins University in 2009. He is a Professor at the Jinan University. His research interests are Econometrics and Labor Economics.
演講摘要 : In this paper, we derive partially identified sets for various nonlinear models with misclassification errors using sup-norm deviations to relax the conditional independence assumptions required for point identification. We express these deviations as a perturbation matrix between an observable matrix and an unobserved eigenvalue-eigenvector decomposition. The perturbation theory of the eigenvalues of a diagonalizable matrix then provides bounds indexed by the upper bound of misreporting probabilities and deviations. As the deviations approach zero, the nonparametric partial identification of the nonlinear models with misclassification error becomes the point identification. We propose a systematic sensitivity analysis to construct the identified sets, incorporating more practical information to determine the upper bounds of deviations. Our simulations imply that the identified sets with the recommended upper bounds can cover the true parameters of interest, and conclusions may apply locally rather than globally. Then, we illustrate the partial identification approach by investigating the impact of misreported schooling on wages.
