A0384
Title: Estimation of threshold dynamic regression for cross-sectional dependent panel time series
Authors: Lulu Wang - University of Southampton (United Kingdom)
Zudi Lu - University of Southampton (United Kingdom) [presenting]
Maria Kyriakou - University of Kent (United Kingdom)
Abstract: The estimation of a family of dynamic threshold models is studied for cross-sectional dependent panel time series. Although the idea of threshold popular in nonlinear analysis for time series has been extended to panel data under cross-sectional independence (CSI), the inference tools built under the CSI, cannot apply to such financial analysis of panel stocks owing to intrinsic cross-sectional dependence (CSD). An estimation of the problem has hence developed under the CSD, with the least squares coefficient estimators shown to be asymptotically normal at a convergence rate of root-nT (with n and T for the sample sizes in cross-section and in time), but their asymptotic variance matrix viably different from that under the CSI. The consistent estimator of the asymptotic variance matrix is further constructed under the CSD. The conditions ensuring the estimators of threshold parameters having a non-standard asymptotic distribution under the CSD are sought in a general setting with the threshold effects diminishing at varied rates in n and T. Monte Carol simulations on the performance under finite sample show that the variances of the estimators would be significantly underestimated, leading to spurious inference, if the panel's cross-sectional dependency were ignored or mistaken to be as the CSI. An empirical application to study the effect of the precipitation on the panel of stocks of FTSE100 confirms that the threshold method facilitates an effective financial analysis.
