EViews.com

Hello, I'm using EViews 13 and trying to use the VARSEL function with the GETS method.

The always-included regressor is just my dependent variable and the list of search-regressors has a constant, the 1-12 lags of the dependent variable and the 1-12 lags of another variable. I'm doing this in a loop for 47 countries.

However, when I run the prg, it pops an error saying "Selection procedure found no variables to include". I tried disabling all the diagnostics tests and setting the terminal condition p-value at 1 to see if it would generate any model, but it continues to show that error for some countries.

The prg line follows below. Any idea what may be causing this issue?

Thanks!

equation tmp_model{%temp}.varsel(method=auto, pval=1, nolm, noarch, nojb, nopet) dp{%temp} @ c dp{%temp}(-1) dp{%temp}(-2) dp{%temp}(-3) dp{%temp}(-4) dp{%temp}(-5) dp{%temp}(-6) dp{%temp}(-7) dp{%temp}(-8) dp{%temp}(-9) dp{%temp}(-10) dp{%temp}(-11) dp{%temp}(-12) _pc96m(-1) _pc96m(-2) _pc96m(-3) _pc96m(-4) _pc96m(-5) _pc96m(-6) _pc96m(-7) _pc96m(-8) _pc96m(-9) _pc96m(-10) _pc96m(-11) _pc96m(-12)

Any chance you could provide the workfile?

Sure, here it goes.

Presumably %temp is changing in some sort of loop - what's it equal to when the program fails?

Depends on the specification I set for GETS. Using EQUATION TMP_MODEL{%TEMP}.VARSEL(METHOD=AUTO, NOJB, NOARCH, NOLM, NOPET, PVAL=0.05), it stops at AUT.

I tried running default settings, and it seemed to go through every country:

If you want to narrow down the selection methods, you'll have to mess around with @errorcount and setmaxerrs to handle cases where the estimation fails and then re-estimate with different settings.

When I run the same model in my PRG it keeps stopping at AUT. The only difference is that I'm computing a rolling RMSE so this is being re-estimated every month in a 120-month rolling estimation window. Maybe this has something to do with the sample? In some cases, it could not be finding a suitable model. Anyway, wasn't this supposed to give me at least 1 model when I disabled all the tests and put the p-value at 1? I thought this way I was removing all the constraints.

The problem is that it is selecting the "empty model" as the best model in your case for these countries (and in the other cases it is selecting the GUM). But the empty model is non-valid.

You could get around it by using the "noempty" option to not include the empty model.

That did the trick! Thanks a lot!