A0924
Title: Multiscale comparison of nonparametric trending coefficients in panel data models
Authors: Marina Khismatullina - Erasmus University Rotterdam (Netherlands) [presenting]
Bernhard van der Sluis - Erasmus University Rotterdam (Netherlands)
Abstract: The purpose is to introduce a novel multiscale testing framework for detecting and localizing slope heterogeneity in panel data models with time-varying coefficients. Unlike traditional global tests, the proposed method not only allows for testing whether regression coefficient functions are homogeneous across cross-sectional units but also identifies specific units and time intervals where these differences occur. The asymptotic validity of the test is established under the general dependence structure. In order to do that, Gaussian strong approximation theory is used together with recent anti-concentration results to construct simultaneous confidence statements over a set of time scales, thereby offering both global and localized inference. Monte Carlo simulations confirm the robust finite sample performance of the method across diverse designs.
