[Gretl-devel] ARMAX and diffs

I was trying to estimate an ARIMA model with a intervention (a level shift). I identificated a ARIMA(0,1,1) model for the data previous to the intervention, so I tried to estimate: Yt=w*St+[(1-\theta L)/(1-L)]*ut St is a step variable, with value 1 after 2002:02 and 0 before that (I know the intervention should be a gradual shift, but this is easier to explain). The problem is that I have no very clear, based on the gretl manual, how gretl treat the difference with the explanatory St variable. So I did some tryings and found what I think is an inconsistency in the treatment of the explanatory variable. Differencing the previous equation I obtain: d_Yt=w*Pt+(1-\theta L)*ut where Pt=(1-L)*St. In the following script I firstly estimate this equation with the data in the attached file. The two following equations estimate the model in levels, but it seems that the gretl native command does not use the St variable as the --x12-arima, because the results are very different. I seems that gretl is not differencing the St variable. I do that manually in the last command and the results now seems consistent with that of X12arima and that obtained with the first command. <script> open import-NE.gdt arima 0 0 1 ; d_Y P 0 arima 0 1 1 ; Y S 0 arima 0 1 1 ; Y S 0 --x-12-arima arima 0 1 1 ; Y P 0 </script> Could you, Allin or Jack, please correct this behaviour? Thanks. -- 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.et.bs.ehu.es