Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Inequality constraint with mata -optimize-


From   Dorothy Bridges <[email protected]>
To   [email protected]
Subject   Re: st: Inequality constraint with mata -optimize-
Date   Mon, 30 May 2011 08:09:48 -0700

Thanks, Stas.  Unfortunately, with the model parameterized as-is, I do
indeed get negative variance estimates (!).  Parameterizing as the
square of something leads to very unexpected (almost certainly wrong)
estimates of the other parameters.  In MATLAB, simply imposing an
inequality constraint seems to "work" .... I realize that there's a
problem with the specification of the model, but it would still be
nice to try something like Cp'>c.

On Sun, May 29, 2011 at 7:44 PM, Stas Kolenikov <[email protected]> wrote:
> The log won't quite do the trick, as it does not allow the variance to
> be zero. Square will; you can parameterize the variance as a square of
> something. However, it would produce (at least) two problems. One is
> that when the true value is at zero, the Jacobian is degenerate
> (there's a row/column corresponding to that square, say variance =
> b*b, of which the derivative is 2*b, and if b is zero, the whole
> column is zero), which screws up both convergence and inference.
> Another problem is that when your model is specified so badly that the
> population "variance" is negative (it does happen in some models,
> although I am not familiar with the model Dorothy mentioned). If this
> happens, the MLE converges to zero producing the above problem, but
> also you may not see the misspecification really occurring as you did
> not allow zero values. So it's a two-edged sword; I would just
> parameterize the model with variance as is, and see if you need to
> worry about negative estimates (for which you still can have
> asymptotically appropriate inference with -robust- standard errors).
>
> On Sun, May 29, 2011 at 7:21 PM, Richard Williams
> <[email protected]> wrote:
>> At 06:09 PM 5/29/2011, Dorothy Bridges wrote:
>>>
>>> Thanks, Stas.  I am using ML to estimate a disequilibrium model, in
>>> which the likelihood function is set up as in Maddala and Nelson,
>>> "Maximum Likelihood Methods for Models of Markets in Disequilibrium,"
>>> Econometrica, 1974.  I simply want to constrain two of the parameters
>>> -- variances of the error terms in the demand and supply equations --
>>> to be greater than or equal to zero.
>>
>> Often this is done by estimating the log of the parameter. See, for example,
>> Stata's -hetprob- program.
>>
>>
>> -------------------------------------------
>> Richard Williams, Notre Dame Dept of Sociology
>> OFFICE: (574)631-6668, (574)631-6463
>> HOME:   (574)289-5227
>> EMAIL:  [email protected]
>> WWW:    http://www.nd.edu/~rwilliam
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/statalist/faq
>> *   http://www.ats.ucla.edu/stat/stata/
>>
>
>
>
> --
> Stas Kolenikov, also found at http://stas.kolenikov.name
> Small print: I use this email account for mailing lists only.
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index