[Gretl-users] forecast error calculation with "--rolling" and 1-step forecasts

Allin Cottrell cottrell at wfu.edu
Sun Nov 15 17:04:09 EST 2015


On Sun, 15 Nov 2015, Sven Schreiber wrote:

> Am 15.11.2015 um 10:41 schrieb Sven Schreiber:
>
>> So does that mean that taking parameter uncertainty into account for a
>> sequence of forecasts has to be done manually with an explicit loop?
>>
>
> Furthermore I found that after "--rolling" I get a not-available error
> if I try to access "$fcerr". (I did omit the varname in the command,
> because according to the doc these accessors can only be used if varname
> is not given.)
>
> So it looks as if --rolling really only has a narrow use case and other
> stuff has to be done manually. Or is it a bug?

It has a narrow use case, as things stand. This option is just not 
very well developed. It's probably fair game for a decent function 
package. If anyone develops such a thing and it's popular, then we 
could think about porting it to C.

Allin Cottrell


More information about the Gretl-users mailing list