A0794
Title: On dynamic threshold modelling of large panel time series adaptive to unknown cross-sectional dependence
Authors: Lulu Wang - Beijing Normal-Hong Kong Baptist University (China) [presenting]
Zudi Lu - City University of Hong Kong (China)
Abstract: Nonlinear large panel time series often exhibit complex cross-sectional dependence (CSD). Inspired by the study of extreme climate effects on a panel of stock return series, the development of a dynamic threshold (auto)regressive panel time series (DyTRPTS) model is proposed, adaptive to the unknown CSD. A consistent estimator of this asymptotic variance matrix is proposed, adaptive to the CSD. Additionally, a non-standard asymptotic distribution for the threshold parameter estimator adaptive to the CSD is established in an extended setting where the threshold effect diminishes at $O(n^{-\alpha_1}T^{-\alpha_2})$, permitting different rates of $\alpha_1, \alpha_2\in[0,\infty)$, rather than in $[0, 1/2)$ required in the literature $(except\alpha_1=\alpha_2=0)$ in n and T, respectively. Interestingly, it is found that the non-standard distributions depend on the convergence of $n^{2\alpha_1-1)}T^{2\alpha_2-1}$ for large n relative to T tending to infinity and on the unknown CSD, differing significantly from those with CSI. Monte Carlo simulations confirm that ignoring or mistaking the panels CSD for CSI leads to underestimated variances of the estimators, causing misleading inferences. Applying the method to study the effect of precipitation on FTSE100 panel stocks in London demonstrates its effectiveness in facilitating accurate analysis.
