Title: Time varying cointegration and Kalman filter
Authors: Burak Eroglu - Bilgi University (Turkey)
Taner Yigit - Bilkent University (Turkey) [presenting]
J Isaac Miller - University of Missouri (United States)
Abstract: Despite its popularity in analyzing different kinds of time varying parameter models, the Kalman Filter (KF) technique has been overlooked so far in the literature on cointegration vector (CV) instability for two reasons: i) because it suggests a very specific (stochastic) form of time variation in the estimated parameters and may fail to capture more general models such as structural breaks, regime switches, etc., and ii) it is vulnerable to the problem of spurious regression because of its Gaussian error assumption and the imposition of a cointegrating relation without the test for it. We propose a method that addresses the latter issue and offers a reliable recovery of the instability in CVs. The results show that KF provides an encompassing estimation strategy to the estimation of CV whether it is time varying or not. Thereby, our methodology proposes a ``universal'' method of estimation for cointegrating vectors, suggests a method of testing for cointegration when the relation is time varying, and avoids the risk of spurious regression.