Hi all, I have a few questions/issues to report with a cointegration example that I am trying to simulate using: nulldata 500 setobs 1 1950 --time-series series u1 = randgen(N,0,1) series u2 = randgen(N,0,1) series Z1 = 0 series Z1 = Z1(-1) + u1 series Z2 = 75 + 0.75*Z1 + u2 1)- When I run a VECM model afterwards, I always have one of the inverse roots on the edge. Am I doing something wrong? 2)- Is there a better way of creating cointegrated series where I can also set the EC coefficients (and/or the lag order)? 3)- Running the EG test, I notice that if the original series are tested with a constant and/or a trend, the residuals in the 4th step are also tested with a constant and/or trend. Can this be a bug, since we are testing residuals? 4)- In the VECM specification window, one can choose up to 10 cointegrating vectors even if there are 2 series. IMHO, it would be nice to put a limit depending on the number of series. Thanks very much in advance for any help or suggestions. Cheers Talha -- “An expert is a person who has made all the mistakes that can be made in a very narrow field.” - Niels Bohr (1885-1962) -- _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

On 07.03.2012 23:18, John C Frain wrote: > 1) I think you will find that > > vecm 1 1 Z2 Z1 --rc > > estimates your system very well.  I dont see the problem. I think Talha means to see unit roots in the output; which is good news, however, since the data were constructed to have them. > > 3) If the series are not cointegrated then the residuals from the > cointegrating regression may have a stochastic trend -- thus the > specification of the test. Hayashi has a very good account of the > Engle Granger cointegration test. No, the (OLS) residuals will have mean zero. (Brownian bridge instead of brownian motion IIRC.) So I tend to agree this could be a bug. BUT: First due to sample shortening because of lags the exact mean-zero property may be lost, which complicates things. And I think the model representation of the residual-based-ADF-test in step 4 may just be wrong, with the result being correct. You can easily check that by manually replicating the steps.

hth, sven _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

2) Consider simulating the model in VECM format. Otherwise I see no problem with your approach.

El 07/03/12 15:56, Talha Yalta escribió: > Hi all, > > I have a few questions/issues to report with a cointegration example > that I am trying to simulate using: > > nulldata 500 > setobs 1 1950 --time-series > series u1 = randgen(N,0,1) > series u2 = randgen(N,0,1) > series Z1 = 0 > series Z1 = Z1(-1) + u1 > series Z2 = 75 + 0.75*Z1 + u2 > > 1)- When I run a VECM model afterwards, I always have one of the > inverse roots on the edge. Am I doing something wrong? If you mean "on the real edge", this is a consequence of the real root z=1 [from polynomial (1_L)] which is present in both series. -- Ignacio Diaz-Emparanza DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA) UPV/EHU Avda. Lehendakari Aguirre, 83 | 48015 BILBAO T.: +34 946013732 | F.: +34 946013754 www.ea3.ehu.es _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users