A0710
Title: Cross-sectionally dependent dynamic threshold regression for nonlinear panel time series: Theory with application
Authors: Maria Kyriacou - University of Kent (United Kingdom) [presenting]
Abstract: The aim is to propose exploring a family of cross-sectionally dependent dynamic threshold models for nonlinear panel analysis. It is motivated by application to the study of the extreme weather effect on the performance of panel stocks that are cross-sectionally dependent. The type of such data, however, exists extensively in applications. The methodology extends the threshold idea in nonlinear time series analysis and also the panel data analysis under cross-sectional independence (CSI) in the literature. The model coefficient estimators are first developed by least squares and are shown to be asymptotically normal at a root-nT convergence rate (with n the sample size in cross section and T in time), but their asymptotic variance matrix is viably different from that under the CSI. A consistent estimator of the asymptotic variance matrix under the CSD is hence suggested. Further, a non-standard asymptotic distribution of the threshold parameter estimator is established in a general setting of the threshold effect diminishing at varied rates in n and T under the CSD. Inference of the threshold effect is also discussed, and an extension to multiple thresholds is made.
