On Wed, 9 Jul 2014, henrique.andrade(a)bb.com.br wrote: > Dear Gretl Community, > > I would like to know how to reproduce an automatic X-12-ARIMA specification > inside Gretl. I'll try to explain better... > > <hansl> > open fedstl.bin > data paynsa > > smpl 2000:01 > </hansl> > > Using the GUI facilities to X-12-ARIMA (Menu -> Variable -> X-12-ARIMA > Analysis) I get the following specification: > > "Final automatic model choice : (0 2 1)(0 1 1)" > > And I get the following model: > > <model> > Estimation converged in 10 ARMA iterations, 31 function evaluations. > ARIMA Model: (0 2 1)(0 1 1) > Nonseasonal differences: 2 > Seasonal differences: 1 > Standard > Parameter Estimate Errors > ----------------------------------------------------- > Nonseasonal MA > Lag 1 0.4834 0.06713 > > Seasonal MA > Lag 12 0.7977 0.05225 > </model> > > So I use it with the "arima" command: > > <hansl> > arima 0 2 1 ; 0 1 1 ; paynsa --nc --x-12-arima > </hansl> > > This gives me the following result: > > <model> > Modelo 3: ARIMA, usando as observações 2000:01-2014:04 (T = 172) > Estimado usando X-12-ARIMA (Máxima verossimilhança exata) > Variável dependente: (1-L)^2(1-Ls) paynsa > > coeficiente erro padrão z p-valor > ---------------------------------------------------------- > theta_1 -0,545388 0,0619664 -8,801 1,35e-018 *** > Theta_1 -0,789758 0,0506616 -15,59 8,66e-055 *** > </model> > > How can I exactly reproduce the model estimated by the X-12-ARIMA procedure?
Note that the "auto" x12a output includes the assessment that the log transformation gives a better fit. So: <hansl> series y = log(paynsa) arima 0 2 1 ; 0 1 1 ; y --nc --x-12-arima </hansl> will replicate the results. Allin Cottrell
