CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: GEL approach for robust regression on spheres Authors:  Fumiya Akashi - University of Tokyo (Japan) [presenting]
Abstract: A nonlinear regression model is considered with a spherical predictor and a possibly heavy-tailed error term. The statistical analysis for spherical and cylindrical data is an important topic in the fields of seismic wave analysis, analysis for orientation of wildfire, wind direction analysis, and so on. We make use of a least absolute deviation-based generalized empirical likelihood (GEL) approach to construct a robust statistic for heavy-tailed observations. The proposed GEL statistics is shown to have a pivotal limit distribution, and based on the asymptotic result, we also propose a method for the construction of confidence intervals for the nonlinear regression function at a certain direction. The finite sample performance of the proposed method is investigated by some simulation experiments.