Forecast averaging

[Katjes]

My forecasts are recursively estimated. Therefore, I think it is reasonable to adjust the training sample for forecast averaging.
Is there an easy command to do this:

Code: Select all

for %weight mse ranks smpl 2013.1 2018.1 GDP.forcavg(name=test_{%weight}, wgttype={%weight},trainsmpl="2010Q1 2012Q4") GDP_f* next

Or do I need to average every period individually with a loop over training sample? e.g. adjusting the training sample for every new forecast quarter?

[EViews Gareth]

I'm not sure I understand the question.

[Katjes]

I now solved the issue with a loop, but was wondering whether there is a more suitable solution:

Code: Select all

'flexible training period" genr mse_fix=NA genr mse_all=NA for !ii=0 to 31 smpl 2011q1+!ii 2011q1+!ii %train_fix="2011q1+"+@str(!ii)+"-8 2011q1+"+@str(!ii)+"-1" ''' fixed evaluation mse sample %train_all="2011q1-1 2011q1+"+@str(!ii)+"-1" 'expanding evaluation mse smpl gdp.forcavg(name=test_a, wgttype=mse, trainsmpl={%train_fix}) gdp_f1 gdp_f2 mse_fix=test_a gdp.forcavg(name=test_b, wgttype=mse, trainsmpl={%train_all}) gdp_f1 gdp_f2 mse_all=test_b next