CMStatistics 2016: Start Registration
View Submission - CFE
Title: Tv-SURE or not Tv-SURE, that is the question: Time-varying coeffcient estimation in SUR models Authors:  Susan Orbe - University of the Basque Country (Spain) [presenting]
Eva Ferreira - University of the Basque Country (Spain)
Isabel Casas - University of Southern Denmark (Denmark)
Abstract: The aim is to study the properties of a kernel estimator for a SURE system with time-varying coeffcients (tv-SURE), under general conditions and also under cross-restrictions. The asymptotical results, consistency and asymptotic normality are obtained in this framework, where the variables are locally stationary rather than stationary. We propose the use of a Kolmogorov-Smirnov statistic to test for cross-restrictions, that allows very general alternatives in a time-varying context. The theoretical results together with the simulation study support the use of tv-SURE in practice. It is important to remark that, contrary to the parametric case,the tv-SURE estimator can be different from the single equation estimator under equal regressors. Since nonparametric estimators are biased, we also need to check that a reduction in the variance does not increase the bias. As an application we estimate the infuence of the Eurostoxx and bond spreads in several domestic indexes returns. Finally we test whether this infuence can be consider the same for some European countries.