GLS under monotone heteroskedasticity
Yoichi Arai,Taisuke Otsu,Mengshan Xu
DOI: https://doi.org/10.1016/j.jeconom.2024.105899
IF: 3.363
2024-11-23
Journal of Econometrics
Abstract:The generalized least square (GLS) is one of the most basic tools in regression analyses. A major issue in implementing the GLS is estimation of the conditional variance function of the error term, which typically requires a restrictive functional form assumption for parametric estimation or smoothing parameters for nonparametric estimation. In this paper, we propose an alternative approach to estimate the conditional variance function under nonparametric monotonicity constraints by utilizing the isotonic regression method. Our GLS estimator is shown to be asymptotically equivalent to the infeasible GLS estimator with knowledge of the conditional error variance, and involves only some tuning to trim boundary observations, not only for point estimation but also for interval estimation or hypothesis testing. Simulation studies and an empirical example illustrate excellent finite sample performances of the proposed method.
economics,social sciences, mathematical methods,mathematics, interdisciplinary applications