Authors: Javier Gomez-Biscarri (Universitat Pompeu Fabra and Barcelona Graduate School of Economics), Javier Hualde (Universidad Pública de Navarra).
Abstract: In cointegration, the I (2) model may be appropriate when the observables are smoother than I (1) and /or when polynomial cointegration is present. Here, the Johansen's maximum likelihood approach has featured prominently. An alternative regression-based methodology has also been proposed, but its application has been mainly restricted to uni-equation systems. In fact, using regression methods to estimate the cointegrating rank and to design correctly cointegrating equations was also not straightforward in the simpler I (1) setting.
This has been addressed by Gomez-Biscarri and Hualde (2015a), and in this paper we extend their approach to the I (2) situation. We propose automatic estimators of the cointegrating rank and of the dimension of a possible cointe-grating subspace in I (2) systems, deriving also data-based just-identifying conditions from which the cointegrating space and subspace can be easily estimated by regression methods. A Monte Carlo analysis of finite sample performance which compares our approach with Johansen's, and an empirical application are also provided.
Link to work
Presentations slides
Abstract: In cointegration, the I (2) model may be appropriate when the observables are smoother than I (1) and /or when polynomial cointegration is present. Here, the Johansen's maximum likelihood approach has featured prominently. An alternative regression-based methodology has also been proposed, but its application has been mainly restricted to uni-equation systems. In fact, using regression methods to estimate the cointegrating rank and to design correctly cointegrating equations was also not straightforward in the simpler I (1) setting.
This has been addressed by Gomez-Biscarri and Hualde (2015a), and in this paper we extend their approach to the I (2) situation. We propose automatic estimators of the cointegrating rank and of the dimension of a possible cointe-grating subspace in I (2) systems, deriving also data-based just-identifying conditions from which the cointegrating space and subspace can be easily estimated by regression methods. A Monte Carlo analysis of finite sample performance which compares our approach with Johansen's, and an empirical application are also provided.
Link to work
Presentations slides