erc/metu
INTERNATIONAL CONFERENCE IN ECONOMICS  IV
September 13-16, 2000, Ankara

Cointegration for Seasonal Time Series Processes

Denise R Osborn (University of Manchester, UK)

Abstract

This paper examines the types of cointegration which can apply between seasonal time series processes. In particular, three types are identified: seasonal cointegration, periodic cointegration and nonperiodic cointegration. The circumstances in which each of these types of cointegration is relevant is examined in a bivariate context. It is shown that, even when both series are first order nonstationary, the possible form(s) for any cointegration depends crucially on the specific univariate unit root properties of the series. When both processes are I(1), cointegration can be only nonperiodic or periodic, with the periodic cointegration coefficients in the latter case related to the periodic integration coefficients of the separate variables. The richest set of cointegration possibilities emerge when both series are seasonally integrated. In this case, all three types of cointegration are possible, while partial cointegration can also hold, whereby stationary linear combinations do not apply to all univariate unit roots. A method is proposed for the nesting of the three types of cointegration. Periodic and seasonal cointegration are shown to imply distinct restrictions, while nonperiodic cointegration is a special case of seasonal cointegration. Thus, testing may be used to discriminate between the possibilities.

Economic Research Center
Middle East Technical University
06531 Ankara Turkey
Phone: + 90 312 210 3044, 210 2003
Fax: +90 312 210 1244
e-mail: metuerc@metu.edu.tr