CMLMT covariance matrix problem

I'm using CMLMT for a maximisation problem, and it goes ok, it finds the maximum, but it only reports standard errors for a couple of variables, while dots show up for the others. It is strange because if in the case of c0.CovParType = 1; I can easily get the standard errors by sqrt(diag(invpd(out.Hessian))) and they are equal to the ones reported for those handful of variables. And if c0.CovParType = 2;, then standard errors are sqrt(diag(invpd(out.Hessian)*out.Xproduct*invpd(out.Hessian))), which are again the same as the ones reported.
If I can calculate them manually for all parameters, why does CMLMT show missing values for some parameters? My likelihood function is pretty simple, no chance for complex numbers, so I don't understand what's going on.

8 Answers


Does your model have constraints?   Standard errors are not available for parameters whose estimates are on constraint boundaries, and missing values are inserted for them.


Also, let me add that the methods you were using to compute the standard errors is incorrect when there are constrained parameters.   Only an approximate covariance matrix is possible in that situation.  Read Section 3.8.1 of the CMLMT 2.0 Manual.   Statistical inference in models with constraints is not a generally resolved issue.  See Section 3.8.2 for some discussion.  I recommend

19. Silvapulle, Mervyn J. and Sen, Pranab K., 2005. Constrained Statistical

Inference. New York: Wiley.

20. Silvapulle, Mervyn J. and Silvapulle, Paramsothy (1995). A score test

against onesided alternatives. Journal of the American Statistical

Association, volume 90, pages 342–349.



No, I have no constraints at all, the model is unconstrained.



You may have some constraints that you aren't aware of, perhaps some bounds on parameters.   I can't think of any other explanation.  If there was a failure of precision or a singular Hessian, a scalar missing value would be returned.  If you would attach to a message your command file and any other files necessary to run it, compressed into a zip file. I'll take a look at it.


Well, I only use one positivity constraint, but I don't impose it as you usually do in CMLMT, i.e. by using c0.C and c0.D. Instead I just take the exponential of the parameter value which ensures positivity. However, that constraint is not binding and CMLMT gives missing values for almost all parameters.

I don't know how to attach file to my message (I didn't find this option), do you have an email address where I could send the file to?

Thanks a lot!



I sent you the files in an email attachment, but I haven't got any reply yet. I am afraid my email ended up in the junk folder, so I ask you here whether you received any email of mine.



I haven't received any files.  I checked and didn't see anything going back to January. 


Thanks a lot for the reply!

I sent my email to the email address you provided above.

You must login to post answers.

Have a Specific Question?

Get a real answer from a real person

Need Support?

Get help from our friendly experts.

Try GAUSS for 14 days for FREE

See what GAUSS can do for your data

© Aptech Systems, Inc. All rights reserved.

Privacy Policy