Re: [Gretl-users] Forecast ARIMA
Am 12.08.2010 19:58, schrieb Allin Cottrell:
On Thu, 12 Aug 2010, [koi8-r] Александр Рогальский wrote:
> During the simulation the following problem occured.
>
> This problem concerns the construction of confidence intervals.
>
> Having developed the ARIMA model (Time Series->ARIMA, using the
> Exact LikelyHood method) I made a forecast of the process in а
> five-year term.
>
> The predicted values were calculated correctly, but all
> confidence intervals had the same size (but the must expand with
> increasing period of pre-emption).
>
> What is the reason of it?
I don't know without seeing your model and your data.
Neither do I, obviously, just a wild guess: Aleksandr says five years
ahead; maybe the confidence interval width converges so quickly at the
beginning and the forecast horizon is so long in comparison that the
convergence process just isn't visible in the graph. But probably that's
not really the problem...
-sven