El Jueves, 21 de Diciembre de 2006 22:23, Javier García escribió:
Before I put a Gretl's example. Data set which I study is monthly. I have
101 observations and, if I apply the Johansen's tests in it, I obtain
The results you obtain may be caused by an inconsistent estimation of the \Pi
matrix because of seasonal unit roots. Gretl does not treat seasonality in
the Johansen test. With monthly data, if you are only interested in the zero
frequency (the trend, or long run) you can:
1) filter your series in this way : X*t = (1+L+L^2+ ...+ L^11)Xt and apply
the Johansen test to the filtered series.
2) Consider at least p=12 (or better 24) lags in the Johansen dialog box.
-Lee, H.S. (1992) "Maximum Likelihood Inferences on Cointegration and seasonal
Cointegration", Journal of Econometrics 54, 1-47.
-Johansen, S. and Schaumburg, E. (1998) "Likelihodd Analysis of Seasonal
Cointegration", Jpournal of Econometrics 88, 301-339.
Dpto. de Economía Aplicada III (Econometría y Estadística)