Page 1 of 1
Maximum Likelihood problem
Posted: Wed Nov 23, 2011 5:53 am
by coldplayer
Uploaded with
ImageShack.us
Re: Maximum Likelihood problem
Posted: Wed Nov 23, 2011 8:31 am
by trubador
EViews does the summation implicitly, so all you have to do is create Ldt and Lst as series and define their equations seperately. And you'll need @dnorm and @cnorm functions to do that. Your likelihood function will then be Log(Ldt + Lst). You can simply use c(1), c(2), ..., c(n) for your unknown parameters or create your own coefficient vectors. I believe, examples in the manual will guide you through the rest...
Re: Maximum Likelihood problem
Posted: Thu Dec 29, 2011 6:38 pm
by coldplayer
Thank you for the answer. However, I've followed your advice but then I've encountered a problem. When I want to estimate a LogL object I get the message
"Missing values in @logl series at current coefficients at observation.....". I've checked all series and none of them have NA in the sample period. What could be the problem? I'm attaching the workfile I'm using.
This is what I wrote in a LogL specification:
Code: Select all
@logl lf
pon=flows-(gama(1)*mmr_eur+gama(2)*yield_ger+gama(3)*ind_eu27+gama(4)*cds_cro+gama(5)*liquid_cro+gama(6)*drate_cro+gama(7)*crobex+gama(8)*resdug_cro)
potr=flows-(beta(1)*cpi_cro+beta(2)*inf_cro+beta(3)*crobex+beta(4)*crbxch+beta(5)*ind_cro+beta(6)*indch_cro+beta(7)*resuvoz_cro+beta(8)*drate_cro+beta(9)*cds_cro)
potraznja=(1/sigma_s(1))*(@dnorm(pon/sigma_s(1)))*(1-(@cnorm(potr/sigma_d(1))))
ponuda=(1/sigma_d(1))*(@dnorm(potr/sigma_d(1)))*(1-(@cnorm(pon/sigma_s(1))))
lf=log(potraznja+ponuda)
What am I doing wrong?
Re: Maximum Likelihood problem
Posted: Thu Dec 29, 2011 7:00 pm
by startz
Check your starting values for the parameters.
Re: Maximum Likelihood problem
Posted: Thu Dec 29, 2011 7:11 pm
by coldplayer
Check your starting values for the parameters.
Starting values are set to 0, but the problem remains if I set them to 1, 10, or any other number.
Re: Maximum Likelihood problem
Posted: Thu Dec 29, 2011 8:57 pm
by startz
set the sigma values to 1000 and it'll run
Re: Maximum Likelihood problem
Posted: Fri Dec 30, 2011 3:11 am
by coldplayer
set the sigma values to 1000 and it'll run
Thanks for the help, it does work!
