| David Harris, University of Melbourne Stephen Leybourne, University of Nottingham Brendan McCabe, University of Liverpool |
| Stochastic Cointegration II: Testing |
| Session: C-9-26 Monday 14 August 2000 by Leybourne, Stephen |
| Harris, McCabe and Leybourne (1999) introduce the concepts of stochastic integration and cointegration where the conventional requirement of co-movement according to an I(0) process is replaced by a weaker concept based on mean reversion. Inference for the stochastically cointegrating vector is discussed. However, before such inference may be used it is necessary to check that the variables under study are in fact stochastically cointegrated, as the results presented do not apply to non-cointegrated systems. This paper shows how to test if a vector of stochastically integrated variables is cointegrated, in either the conventional or stochastic senses, against the alternative of no cointegration. If the variables are found to be cointegrated a supplementary test is used to distinguish between conventional and stochastic cointegration. Some Monte Carlo studies are presented to assess the accuracy of the size of the tests and the power properties of the tests are also examined. An application to Purchasing Power Parity data is given as an empirical example. |