stroring starting values

[KarinL]

Hello, I'm using Eviews7.
I am currently doing some MLE estimation using a logl object. Apparently convregence depends strongly on the starting values of the parameters. In programming the estimation procedure, I'm using the statement parm c(1) 0 c(2) 1 etc.
I'd like to also store the starting values I'm using in my output. Any ideas how I can handle that? Thanks.

[EViews Gareth]

The LOGL estimation command has a showopt option that will print the starting values out in the estimation output.

[KarinL]

I'm trying to solve my LOGL object consecutively for differing starting values right now. Is there a command I could use in my program, telling to delete an LOGL object, if it does not converge? (or if there comes the warning - near singular cov matrix and I get NAs for standard errors). This would be really helpful in automating the search...

[EViews Gareth]

Unfortunately there is no way to get that information back, other than freezing the estimation output into a table, then searching for the terms you want to test.

[EViews Glenn]

You can test whether the standard errors of the coefficients are [edit] NA and then conditionally delete the object

[KarinL]

Is there a way to obtain e.g. a standard error of one of the coefficients estimated in the logl?

If so, I could use an if-statement of the form
if SE = "NA" delete LOGL

would that be possible?

[EViews Gareth]

Yes, although that won't help you with convergence not achieved.

Code: Select all

if logl.@stderrs(1) = na then d logl endif

[KarinL]

thanks! That helps - if I let the program run for different starting values I won't have to sort out the ones that didn't converge by hand ...

[KarinL]

Is there a way to get access to the p-value of a coefficient? I'd like to automatically delete estimation results with a very bad p-value.
I didn't find anything in the manual... thanks!

[EViews Gareth]

Nope, they have to be computed manually from the t-stat.