R help - Mar 2011 - recommended reading for manual maximum likelihood of a system of equations

Hi everyone,

I am looking for some recommended reading.

I want to read up on the estimation systems of linear equations using
maximum likelihood?
I have a strongly applied bias, I want to be able to do such estimation myself.
Reading with examples would be great.
Something which also works through a concentrated maximum likelihood
estimation would be even better!

I want to use a likelihood ratio test to compare some nested models of
linear simultaneous equations.
I have estimated the systems using   systemfit()   and
nlsystemfit().  systemfit()   can be used with   lrtest()   which
automatically obtains the log-likelihood values.
Unfortunately one of my models requires nonlinear coefficient
restrictions and   lrtest()   does not seem to be usable with
nlsystemfit().

I decided it would be a good idea to program the likelihood estimation myself.
This would solve the problem and would also be a great learning experience.

All suggestions will be highly appreciated!

Thanks a lot,

Alex Olssen
Motu Economic and Public Policy Research
Wellington
New Zealand

Hi Alex

On 24 March 2011 04:47, Alex Olssen <alex.olssen at gmail.com>
wrote:
> I am looking for some recommended reading.
>
> I want to read up on the estimation systems of linear equations using
> maximum likelihood?
For a comprehensive review of estimating systems of linear equations, see:
   http://www.jstatsoft.org/v23/i04/
or
   Greene, Econometric Analysis, 6th ed., 2008, chapter 10.

For ML estimation in R, see:
   http://www.springerlink.com/content/973512476428614p/

For ML estimations of systems of linear equations, see:
   Greene, Econometric Analysis, 6th ed., 2008, section 13.6.2.

Please note that the iterated SUR estimator should converge to the ML estimator.
> I have a strongly applied bias, I want to be able to do such estimation
myself.
> Reading with examples would be great.
> Something which also works through a concentrated maximum likelihood
> estimation would be even better!
>
> I want to use a likelihood ratio test to compare some nested models of
> linear simultaneous equations.
> I have estimated the systems using ? systemfit() ? and
> nlsystemfit(). ?systemfit() ? can be used with ? lrtest() ? which
> automatically obtains the log-likelihood values.
> Unfortunately one of my models requires nonlinear coefficient
> restrictions and ? lrtest() ? does not seem to be usable with
> nlsystemfit().
Yes. Please note: in contrast to systemfit(), the function
nlsystemfit() is still under development; it has convergence problems
rather often and only a few methods have been implemented yet.
> I decided it would be a good idea to program the likelihood estimation
myself.
> This would solve the problem and would also be a great learning experience.
You are invited to implement the ML estimation in the "systemfit"
package. You could join the developer team at R-Forge.

Best wishes from Copenhagen,
Arne

-- 
Arne Henningsen
http://www.arne-henningsen.name