Page 1 of 1
Programming a pseudo out of sample test with mixed frequencies
Posted: Mon Jun 17, 2019 7:06 am
by miorinnovo
Is it possible to run a program that will calculate the rmse for a quarterly model with a monthly independent variable after each month of data is received. In other words. I have a simple quarterly model with one independent variable which is actually a monthly frequency converted to quarterly.
I know how to do the pseudo out of sample rmse test if i just assume the monthly series is quarterly, but Id actually like to know how well the model does with only 0,1 or 2 months of the independent series.
Is this type of thing possible?
Re: Programming a pseudo out of sample test with mixed frequencies
Posted: Mon Jun 17, 2019 8:05 am
by EViews Gareth
Yes.
Re: Programming a pseudo out of sample test with mixed frequencies
Posted: Mon Jun 17, 2019 8:13 am
by miorinnovo
Yes.
Great. Could you direct me to some information on how to do it?
Re: Programming a pseudo out of sample test with mixed frequencies
Posted: Mon Jun 17, 2019 8:19 am
by EViews Gareth
It is relatively straight forward, but laborious. I'm not sure we have concrete examples, other than this blog post:
http://blog.eviews.com/2018/12/nowcasti ... basis.html
which has layers of complexity above what you want, but if you can follow Michael's program, writing your own should be simple.
Re: Programming a pseudo out of sample test with mixed frequencies
Posted: Mon Jun 17, 2019 11:46 am
by miorinnovo
I'm not sure this is what I'm looking for. Essentially I just want to know how to calculate the rmse for different time horizons where those time horizons reflect how many months of data I have. SO for my x, I would extend the series by using a moving average to the end of a quarter whenever there is missing data. I'd like to know what the RMSE is when 1 month is missing and filled in, when 2 months are missing and filled in, when 3 months... etc.
If anyone has any pointers it would be appreciated.