[Gretl-users] ARIMA forecast (very) strange behavior

Allin Cottrell cottrell at wfu.edu
Tue Sep 30 17:51:54 EDT 2014


On Tue, 30 Sep 2014, henrique.andrade at bb.com.br wrote:

> Dear Gretl Team,
> 
> Please take a look at the following (very simple) script:
> 
> <hansl>
> open "Exemplo.gdt"
> 
> smpl 2004:10 2011:10
> 
> arima 0 0 1 ; 1 0 0 ; ipc_dive_m --x-12-arima
> fcast 2011:11 2011:11
> 
> <hansl>
> 
> It gives me infinite values for the sum of squared residuals and a very
> strange behaviour of the forecasted value (9,87378e+307). The ipc_dive_m is
> an inflation measure from Brazil.

Confirmed here. The strange numbers are obviously the result of an 
NA value getting involved in the calculations where this was not 
expected and therefore not handled properly. The next question is 
how/why did the NA value arise?

As Hélio has remarked, this doesn't happen when the model is 
estimated via gretl's native arima functionality. So it's something 
to do with our interface with X-12-ARIMA (or X13-ARIMA-SEATS). I'm 
working on it.

Allin Cottrell


More information about the Gretl-users mailing list