Dear all, I've got a problem with fcast() using it within a bundle context. In the example script below, I want to set up a rolling (both sample start and sample end are moving forward at each iteration) forecast exercise. However, I obtain an issue with the "--dynamic" option. It is about the line "fcast ($t2+1) ($t2+b.k) --dynamic". You could even try "fcast --dynamic". Surprisingly, the rolling option works: "fcast ($t2+1) ($t2+b.k) b.k --rolling". I also set up an alternative function which doesn't make use of a bundle, and there it works fine (see the attached file please). Any ideas what's wrong here?

Hi Sven, let me explain in more detail what I tried to program: - Send a dataset comprising the first T of total T+k observations to function F where k>0 - Restrict the dataset to t=1..(T-h) observations within function F using the "smpl" command - Forecast the remaining h observations dynamically This should be doable (I guess) as I just want to access the T observations which I initially sent to F. The --rolling option seems to recognize/accept this but for some reason not the --dynamic option. I just saw that the 2nd illustrative example -- the one which works but does not involve bundles -- can be simplified as follows: <hansl> set verbose off open denmark.gdt -q series x = diff(LRY) function void dynfc (series x, int R, int k) smpl --no-missing x scalar t1 = $t1 scalar nsel = t1+(R-2) scalar n = nobs(x) scalar nstep = (n-R-k) loop i=1..nstep -q if i==1 smpl t1 nsel endif # fit to the training test ols x 0 x(-1 to -2) # OoS forecast fcast ($t2+1) ($t2+3) --dynamic smpl +1 +1 # roll forward endloop end function dynfc(x, 10,1) </hansl> Artur Am 23.01.2017 um 13:41 schrieb Sven Schreiber: > Am 23.01.2017 um 09:09 schrieb Artur Tarassow: >> Dear all, >> >> I've got a problem with fcast() using it within a bundle context. In the >> example script below, I want to set up a rolling (both sample start and >> sample end are moving forward at each iteration) forecast exercise. >> However, I obtain an issue with the "--dynamic" option. >> >> It is about the line "fcast ($t2+1) ($t2+b.k) --dynamic". You could even >> try "fcast --dynamic". Surprisingly, the rolling option works: "fcast >> ($t2+1) ($t2+b.k) b.k --rolling". >> >> I also set up an alternative function which doesn't make use of a >> bundle, and there it works fine (see the attached file please). >> >> Any ideas what's wrong here? >> > > Artur, I find these functions hard to verify in terms of the samples > they're using. A function typically wouldn't have access to > out-of-sample observations (with respect to the active sample at the > time the function is called). It looks as if that could be an issue, > depending on the values of $t2/$t1/R. Maybe you could add some explicit > error checks about that. > However, I agree that it's surprising that --rolling would work if > --dynamic doesn't. So no, don't have very concrete ideas. > > good luck, > sven > _______________________________________________ > Gretl-users mailing list > Gretl-users(a)lists.wfu.edu > http://lists.wfu.edu/mailman/listinfo/gretl-users

Sorry, for writing again. But I should have provided a more tidy example. Please find attached the two approaches: 1) no bundle involved and 2) bundle involved computing the rolling forecast of a model.

Dear all, I spent quite some time on this issue now but still can't figure out what causes the problem. I wrote two versions of a function for conducting a rolling DYNAMIC forecast analysis. I attached the script.

The error message when calling tscvrol() is: <output> --dynamic: inapplicable option *** error within loop in function tscvrol > fcast ($t2+1) ($t2+b.nhor) --dynamic

Sorry, that was not a good suggestion on my part. String substitution is generally a last resort and it's not needed in this case, since commands are supposed to accept scalar variables in place of numerical constants. My revised recommendation here would be: scalar ft1 = $t2 + 1 scalar ft2 = $t2 + b.nhor fcast ft1 ft2 --dynamic I gather that your original version, fcast ($t2+1) ($t2+b.nhor) --dynamic worked OK once you sorted out the "dynamic" issue, but you "got lucky" in that respect. Such expressions are fragile in a command context. For example if you inserted spaces fcast ($t2 + 1) ... the parsing would fall apart. (And I don't consider that a bug, because commands generally expect space-separated arguments.) Allin _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Am 24.01.2017 um 12:53 schrieb Artur T.: > The error message when calling tscvrol() is: > > <output> > --dynamic: inapplicable option Actually I think gretl is probably right: In your second example gretl has no way of knowing that the xlist contains the lags of your y variable. Because you have transformed your x input into a matrix, and then arbitrarily back to a series. How is the function tscvrol supposed to know how you constructed y previously? cheers, sven

Commands (as opposed to functions) are not generally equipped to

string ft1 = sprintf("%d", $t2+1) string ft2 = sprintf("%d", $t2+b.nhor) fcast @ft1 @ft2 --dynamic Allin