Re: [Gretl-users] Wrong Coefficient Estimation for ARIMA
On 20/06/12 11:47, Sanzad Siddique wrote:
Hi,
I have tried the ARIMA command in GRETL with the attached time series
data. The time series is constructed with the below formula ( a- is
normal random values):
z(t) = 0.50 + 0.50*z(t-1) + 0.40*z(t-2) + a(t)-0.25*a(t-1) -0.75*a(t-2);
The AR roots are approx -2.33 and 1.07 and the MA roots are -1.33 and 1.
The model is non-invertible and almost non-stationary
and one of the AR roots (1.07) practically cancells one of the MA roots (1).
I think gretl's arima engine restricts itself to an invertible area.
--
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.ehu.es/ignacio.diaz-emparanza/
www.ea3.ehu.es